§
    l¬¸jcƒ ã                  óV  — d dl mZ d dlZd dlZd dlZd dlZd dlZd dlmZ d dl	m
Z
 d dlZd dlZd dlmZ d dlmZ d dlmZmZmZmZmZmZmZmZmZmZmZmZmZmZm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z)m*Z*m+Z+m,Z,m-Z-m.Z.m/Z/m0Z0m1Z1m2Z2m3Z3m4Z4m5Z5m6Z6m7Z7m8Z8m9Z9m:Z:m;Z;m<Z<m=Z=m>Z>m?Z?m@Z@mAZAmBZBmCZCmDZDmEZEmFZFmGZGmHZHmIZImJZJmKZKmLZLmMZMmNZNmOZO d dlPmQZQmRZRmSZSmTZTmUZUmVZVmWZWmXZXmYZY d d	lZm[Z[m\Z\m]Z]m^Z^m_Z_ d d
l`maZambZbmcZcmdZdmeZemfZfmgZgmhZhmiZi d dljmkZk d dllmmZm d dlnmoZompZpmqZqmrZrmsZsmtZtmuZumvZvmwZwmxZx  G d„ dej        jy        ¦  «        Zzd„ Z{ G d„ dez¦  «        Z| G d„ de|¦  «        Z} G d„ de|¦  «        Z~ G d„ dez¦  «        Z G d„ de|¦  «        Z€ G d„ de|¦  «        Z� G d„ dez¦  «        Z‚ G d„ d e|¦  «        Zƒ G d!„ d"e|¦  «        Z„ G d#„ d$e|¦  «        Z… G d%„ d&e…e~¦  «        Z† G d'„ d(ej        j‡        ¦  «        Zˆ ej‰        ¦   «         ZŠej‹        d4d)„¦   «         ZŒ G d*„ d+¦  «        Z�d5d,„ZŽd6d/„Z�d7d0„Z�d1„ Z‘d2„ Z’d3„ Z“dS )8é    )ÚannotationsN)Údeepcopy)ÚPath)Únn)Úcheck_class_names)>ÚAIFIÚC1ÚC2ÚC2PSAÚC3ÚC3TRÚELAN1ÚOBBÚOBB26ÚPSAÚSPPÚSPPELANÚSPPFÚA2C2fÚAConvÚADownÚ
BottleneckÚBottleneckCSPÚC2fÚC2fAttnÚC2fCIBÚC2fPSAÚC3GhostÚC3k2ÚC3xÚCBFuseÚCBLinearÚClassifyÚConcatÚConvÚConv2ÚConvTransposeÚDepthÚDetectÚDWConvÚDWConvTranspose2dÚFocusÚGhostBottleneckÚ	GhostConvÚHGBlockÚHGStemÚImagePoolingAttnÚIndexÚLRPCHeadÚPoseÚPose26ÚRepC3ÚRepConvÚRepNCSPELAN4ÚRepVGGDWÚResNetLayerÚRTDETRDecoderÚSCDownÚSegmentÚ	Segment26ÚSemanticSegmentÚTorchVisionÚWorldDetectÚYOLOEDetectÚYOLOESegmentÚYOLOESegment26Ú	v10Detect)	ÚDEFAULT_CFG_DICTÚLOGGERÚ	SAFE_LOADÚSETTINGSÚWINDOWSÚYAMLÚIterableSimpleNamespaceÚcolorstrÚemojis)ÚREMOTE_FILE_PREFIXESÚ
check_fileÚcheck_requirementsÚcheck_suffixÚ
check_yaml)	ÚDepthLoss26ÚE2ELossÚ
PoseLoss26ÚSemanticSegmentationLossÚv8ClassificationLossÚv8DetectionLossÚ	v8OBBLossÚ
v8PoseLossÚv8SegmentationLoss)Úmake_divisible)Ú
torch_load)
Úfuse_conv_and_bnÚfuse_deconv_and_bnÚinitialize_weightsÚintersect_dictsÚis_qatÚ
model_infoÚrestore_qatÚ	scale_imgÚsmart_inference_modeÚ	time_syncc                  óz   ‡ — e Zd ZdZd„ Zdd„Zdd„Zd„ Zd„ Zdd„Z	d„ Z
dd„Zˆ fd„Zdd„Zddd„Zdd„Zd„ Zˆ xZS ) Ú	BaseModela  Base class for all YOLO models in the Ultralytics family.

    This class provides common functionality for YOLO models including forward pass handling, model fusion, information
    display, and weight loading capabilities.

    Attributes:
        model (torch.nn.Sequential): The neural network model.
        save (list): List of layer indices to save outputs from.
        stride (torch.Tensor): Model stride values.

    Methods:
        forward: Perform forward pass for training or inference.
        predict: Perform inference on input tensor.
        fuse: Fuse Conv/BatchNorm layers and reparameterize for optimization.
        info: Print model information.
        load: Load weights into the model.
        loss: Compute loss for training.

    Examples:
        Create a BaseModel instance
        >>> model = BaseModel()
        >>> model.info()  # Display model information
    c                óp   — t          |t          ¦  «        r | j        |g|¢R i |¤ŽS  | j        |g|¢R i |¤ŽS )a'  Perform forward pass of the model for either training or inference.

        If x is a dict, calculates and returns the loss for training. Otherwise, returns predictions for inference.

        Args:
            x (torch.Tensor | dict): Input tensor for inference, or dict with image tensor and labels for training.
            *args (Any): Variable length argument list.
            **kwargs (Any): Arbitrary keyword arguments.

        Returns:
            (torch.Tensor): Loss if x is a dict (training), or network predictions (inference).
        )Ú
isinstanceÚdictÚlossÚpredict)ÚselfÚxÚargsÚkwargss       úd/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/ultralytics/nn/tasks.pyÚforwardzBaseModel.forward’   sZ   € õ �a�ÑÔð 	1Ø�4”9˜QÐ0 Ð0Ð0Ð0¨Ð0Ð0Ð0ØˆtŒ|˜AÐ/ Ð/Ð/Ð/¨Ð/Ð/Ð/ó    FNc                ó^   — |r|                       |¦  «        S |                      |||¦  «        S )a®  Perform a forward pass through the network.

        Args:
            x (torch.Tensor): The input tensor to the model.
            profile (bool): Print the computation time of each layer if True.
            augment (bool): Augment image during prediction.
            embed (list, optional): A list of layer indices to return embeddings from.

        Returns:
            (torch.Tensor): The last output of the model.
        )Ú_predict_augmentÚ_predict_once)rp   rq   ÚprofileÚaugmentÚembeds        rt   ro   zBaseModel.predict£   s8   € ð ð 	,Ø×(Ò(¨Ñ+Ô+Ð+Ø×!Ò! ! W¨eÑ4Ô4Ð4rv   c                óÎ  ‡‡— g g g cŠ}}|rt          |¦  «        ndh}t          |¦  «        }| j        D �]/}|j        dk    r;t	          |j        t
          ¦  «        r‰|j                 nˆˆfd„|j        D ¦   «         Š|r|                      |‰|¦  «          |‰¦  «        Š‰                     |j        | j	        v r‰nd¦  «         |j        |v r”|                     t          j        j                             ‰d¦  «                             d¦  «                             d¦  «        ¦  «         |j        |k    r+t          j        t          j        |d¦  «        d¬¦  «        c S �Œ1‰S )aq  Perform a forward pass through the network.

        Args:
            x (torch.Tensor): The input tensor to the model.
            profile (bool): Print the computation time of each layer if True.
            embed (list, optional): A list of layer indices to return embeddings from.

        Returns:
            (torch.Tensor): The last output of the model.
        éÿÿÿÿc                ó0   •— g | ]}|d k    r‰n‰|         ‘ŒS ©r~   © ©Ú.0Újrq   Úys     €€rt   ú
<listcomp>z+BaseModel._predict_once.<locals>.<listcomp>Ã   ó*   ø€ Ð8]Ð8]Ð8]ÐTU¸aÀ2ºg¸g¸¸È1ÈQÌ4Ð8]Ð8]Ð8]rv   N©é   r‰   r‰   r   ©Údim©Ú	frozensetÚmaxÚmodelÚfrl   ÚintÚ_profile_one_layerÚappendÚiÚsaveÚtorchr   Ú
functionalÚadaptive_avg_pool2dÚsqueezeÚunbindÚcat)	rp   rq   rz   r|   ÚdtÚ
embeddingsÚmax_idxÚmr…   s	    `      @rt   ry   zBaseModel._predict_once³   so  øø€ ð   BÐˆˆ2ˆzØ$)Ð3•	˜%Ñ Ô Ð °¨tˆÝ�e‘*”*ˆØ”ð 
	Iñ 
	IˆAØŒs�bŠyˆyÝ(¨¬­cÑ2Ô2Ð]�A�a”c”F�FÐ8]Ð8]Ð8]Ð8]Ð8]ÐYZÔY\Ð8]Ñ8]Ô8]�Øð 2Ø×'Ò'¨¨1¨bÑ1Ô1Ð1Ø��!‘”ˆAØ�HŠH˜!œ# ¤Ð*Ð*�Q�Q°Ñ5Ô5Ð5ØŒs�eˆ|ˆ|Ø×!Ò!¥%¤(Ô"5×"IÒ"IÈ!ÈVÑ"TÔ"T×"\Ò"\Ð]_Ñ"`Ô"`×"hÒ"hÐikÑ"lÔ"lÑmÔmÐmØ”3˜'’>�>Ý œ<­¬	°*¸aÑ(@Ô(@ÀaÐHÑHÔHÐHÐHÐHùØˆrv   c                ón   — t          j        | j        j        › d�¦  «         |                      |¦  «        S )zFPerform augmentations on input image x and return augmented inference.zR does not support 'augment=True' prediction. Reverting to single-scale prediction.)rG   ÚwarningÚ	__class__Ú__name__ry   )rp   rq   s     rt   rx   zBaseModel._predict_augmentÎ   sC   € åŒØŒ~Ô&ð 5ð 5ð 5ñ	
ô 	
ð 	
ð ×!Ò! !Ñ$Ô$Ð$rv   c           	     óH  — 	 ddl }n# t          $ r d}Y nw xY w|| j        d         k    ot          |t          ¦  «        }|r;|                     ||r|                     ¦   «         n|gd¬¦  «        d         dz  dz  nd}t          |                      ¦   «         ¦  «        j	        }t          |¦  «        }t          d¦  «        D ]#}	 ||r|                     ¦   «         n|¦  «         Œ$|                     t          |¦  «        |z
  d	z  ¦  «         || j        d         k    r t          j        d
d›ddd›ddd›d�¦  «         t          j        |d         d›d|d›d|j        d›d|j        › �¦  «         |r/t          j        t#          |¦  «        d›ddd›ddd›d�¦  «         dS dS )a/  Profile the computation time and FLOPs of a single layer of the model on a given input.

        Args:
            m (torch.nn.Module): The layer to be profiled.
            x (torch.Tensor): The input data to the layer.
            dt (list): A list to store the computation time of the layer.
        r   Nr~   F)ÚinputsÚverboseg    eÍÍAé   é
   éd   z	time (ms)z>10sú ÚGFLOPsÚparamsz  modulez10.2fú10.0fú  ú-z  Total)ÚthopÚImportErrorr�   rl   Úlistrz   ÚcopyÚnextÚ
parametersÚdevicerh   Úranger“   rG   ÚinfoÚnpÚtypeÚsum)
rp   rŸ   rq   rœ   r°   ÚcÚflopsr¶   ÚtÚ_s
             rt   r’   zBaseModel._profile_one_layerÖ   s÷  € ð	ØˆKˆKˆKˆKøÝð 	ð 	ð 	ØˆDˆDˆDð	øøøð �”˜B”ÒÐ7¥J¨qµ$Ñ$7Ô$7ˆØ^bÐi�—’˜Q°AÐ(<¨¯ª©¬¨¸1Ð'=Àu�ÑMÔMÈaÔPÐSVÑVÐYZÑZÐZÐhiˆÝ�d—o’oÑ'Ô'Ñ(Ô(Ô/ˆÝ�fÑÔˆÝ�r‘”ð 	$ð 	$ˆAØˆA˜!Ð"ˆa�fŠf‰hŒhˆh Ñ#Ô#Ð#Ð#Ø
�	Š	•9˜VÑ$Ô$ qÑ(¨CÑ/Ñ0Ô0Ð0Ø�”
˜1”ÒÐÝŒK˜;ÐUÐUÐU¨hÐUÐUÐU¸hÐUÐUÐUÐUÑVÔVÐVÝŒ�r˜"”vÐJÐJÐJ eÐJÐJÐJ°A´DÐJÐJÐJÀ!Ä&ÐJÐJÑKÔKÐKØð 	IÝŒK�3˜r™7œ7ÐGÐGÐG¨3ÐGÐGÐG°cÐGÐGÐGÐGÑHÔHÐHÐHÐHð	Ið 	Is   ‚ ‡•Té€  c                ó$  — t          | ¦  «        rt          t          t          t          fnd}|                      ¦   «         �sÏ| j                             ¦   «         D �]�}t          ||¦  «        rŒt          |t          t          t          f¦  «        rtt          |d¦  «        rdt          |t          ¦  «        r|                     ¦   «          t          |j        |j        ¦  «        |_        t!          |d¦  «         |j        |_        t          |t          ¦  «        rKt          |d¦  «        r;t'          |j        |j        ¦  «        |_        t!          |d¦  «         |j        |_        t          |t          ¦  «        r |                     ¦   «          |j        |_        t          |t          ¦  «        r |                     ¦   «          |j        |_        t          |t,          ¦  «        r|                     ¦   «          �ŒŸ|                      ||¬¦  «         | S )aj  Fuse Conv/ConvTranspose and BatchNorm layers, and reparameterize RepConv/RepVGGDW for improved efficiency.

        Args:
            verbose (bool): Whether to print model information after fusion.
            imgsz (int | list): Input image size used for FLOPs calculation.

        Returns:
            (torch.nn.Module): The fused model is returned.
        r�   Úbn)r¦   Úimgsz)rc   r&   r'   r7   r9   Úis_fusedr�   Úmodulesrl   r%   r*   ÚhasattrÚ
fuse_convsr_   ÚconvrÂ   ÚdelattrÚforward_fuseru   r`   Úconv_transposeÚfuser)   r¸   )rp   r¦   rÃ   ÚskiprŸ   s        rt   rÌ   zBaseModel.fuseð   sº  € õ =CÀ4¹L¼LÐP••}¥g­xÐ8Ð8ÈbˆØ�}Š}‰Œñ 	4Ø”Z×'Ò'Ñ)Ô)ð ñ �Ý˜a Ñ&Ô&ð ØÝ˜a¥$­­vÐ!6Ñ7Ô7ð /½GÀAÀtÑ<LÔ<Lð /Ý! !¥UÑ+Ô+ð 'ØŸš™œ˜Ý-¨a¬f°a´dÑ;Ô;�A”FÝ˜A˜tÑ$Ô$Ð$Ø !¤�A”IÝ˜a¥Ñ/Ô/ð /µG¸A¸tÑ4DÔ4Dð /Ý'9¸!Ô:JÈAÌDÑ'QÔ'Q�AÔ$Ý˜A˜tÑ$Ô$Ð$Ø !¤�A”IÝ˜a¥Ñ)Ô)ð /Ø—L’L‘N”N�NØ !¤�A”IÝ˜a¥Ñ*Ô*ð /Ø—F’F‘H”H�HØ !¤�A”IÝ˜a¥Ñ(Ô(ð Ø—F’F‘H”H�HùØ�IŠI˜g¨UˆIÑ3Ô3Ð3àˆrv   c                óZ   — t          d„ |                      ¦   «         D ¦   «         ¦  «         S )z/Return True once fuse() has nothing left to do.c              3  ó(  K  — | ]�}t          |t          t          f¦  «        rt          |d ¦  «        p\t          |t          t
          f¦  «        ot          |d¦  «        p0t          |t          ¦  «        o|j        duot          |dd¦  «        duV — ŒŽdS )rÂ   Úconv1NÚone2one_cv2)	rl   r%   r'   rÆ   r7   r9   r)   Úcv2Úgetattr)rƒ   rŸ   s     rt   ú	<genexpr>z%BaseModel.is_fused.<locals>.<genexpr>  s´   è è € ð 
ð 
ð õ ˜�D¥-Ð0Ñ1Ô1ÐFµg¸aÀÑ6FÔ6Fð mÝ˜1�w­Ð1Ñ2Ô2ÐJµw¸qÀ'Ñ7JÔ7Jðmå˜1�fÑ%Ô%Ðk¨!¬%°tÐ*;ÐkÅÈÈ=ÐZ^Ñ@_Ô@_ÐgkÐ@kð
ð 
ð 
ð 
ð 
ð 
rv   )ÚanyrÅ   ©rp   s    rt   rÄ   zBaseModel.is_fused  s?   € åð 
ð 
ð —\’\‘^”^ð	
ñ 
ô 
ñ 
ô 
ð 
ð 	
rv   c                ó(   — t          | |||¬¦  «        S )a#  Print model information.

        Args:
            detailed (bool): If True, prints out detailed information about the model.
            verbose (bool): If True, prints out the model information.
            imgsz (int): The size of the image used for computing model information.
        )Údetailedr¦   rÃ   )rd   )rp   rØ   r¦   rÃ   s       rt   r¸   zBaseModel.info   s   € õ ˜$¨¸7È%ÐPÑPÔPÐPrv   c                óŒ  •— t          ¦   «                              |¦  «         | j        d         }t          |t          ¦  «        r@ ||j        ¦  «        |_         ||j        ¦  «        |_         ||j        ¦  «        |_        n?t          |t          ¦  «        r* ||j        ¦  «        |_         ||j	        ¦  «        |_	        | S )a  Apply a function to all tensors in the model, including Detect head attributes like stride and anchors.

        Args:
            fn (function): The function to apply to the model.

        Returns:
            (BaseModel): An updated BaseModel object.
        r~   )
ÚsuperÚ_applyr�   rl   r)   ÚstrideÚanchorsÚstridesr;   Ú
valid_mask©rp   ÚfnrŸ   r¢   s      €rt   rÛ   zBaseModel._apply*  s¯   ø€ õ 	‰Œ�Š�rÑÔÐØŒJ�rŒNˆÝØ�vñ
ô 
ð 	,ð �r˜!œ(‘|”|ˆAŒHØ˜˜1œ9™œˆAŒIØ˜˜1œ9™œˆAŒIˆIÝ˜�=Ñ)Ô)ð 	,Ø˜˜1œ9™œˆAŒIØ˜2˜aœlÑ+Ô+ˆAŒLØˆrv   c           	     óê  — t          |t          ¦  «        r|                     d¦  «        p|d         n|}|                     ¦   «                              ¦   «         }|                      |||¬¦  «        }t          ||                      ¦   «         ¦  «        }|                      |d¬¦  «         t          |¦  «        |z   }d}|                      ¦   «         }	||vr|||	v rx|	|         j	        \  }
}}}||         j	        \  }}}}||k    rN||k    rHt          |
|¦  «        t          ||¦  «        }}
||         d|
…d|…f         |	|         d|
…d|…f<   |dz  }t          |d	d¦  «        | _        |rƒt          j        d
|› dt          | j                             ¦   «         ¦  «        › d�¦  «          t          |dd„ ¦  «        ¦   «         r,|                      ¦   «         st          j        d¦  «         dS dS dS dS )zÔLoad weights into the model.

        Args:
            weights (dict | torch.nn.Module): The pre-trained weights to be loaded.
            verbose (bool, optional): Whether to log the transfer progress.
        Úemar�   )r¦   F)Ústrictzmodel.0.conv.weightNr‰   Úpt_pathzTransferred ú/z items from pretrained weightsrÄ   c                 ó   — dS )NFr�   r�   rv   rt   ú<lambda>z BaseModel.load.<locals>.<lambda>]  s   € °%€ rv   zVPretrained weights are fused for inference; train from the unfused checkpoint instead.)rl   rm   ÚgetÚfloatÚ
state_dictÚ_remap_cls_by_namesrb   Úload_state_dictÚlenÚshapeÚminrÓ   rå   rG   r¸   r�   rÄ   r¡   )rp   Úweightsr¦   r�   ÚcsdÚcls_remappedÚupdated_csdÚlen_updated_csdÚ
first_convrë   Úc1Úc2ÚhÚwÚcc1Úcc2ÚchÚcws                     rt   ÚloadzBaseModel.load@  s$  € õ =GÀwÕPTÑ<UÔ<UÐb�—’˜UÑ#Ô#Ð7 w¨wÔ'7øÐ[bˆØ�kŠk‰mŒm×&Ò&Ñ(Ô(ˆð ×/Ò/°°UÀGÐ/ÑLÔLˆå% c¨4¯?ª?Ñ+<Ô+<Ñ=Ô=ˆØ×Ò˜[°ÐÑ7Ô7Ð7Ý˜kÑ*Ô*¨\Ñ9ˆØ*ˆ
à—_’_Ñ&Ô&ˆ
Ø˜[Ð(Ð(¨Z¸:Ð-EÐ-EØ% jÔ1Ô7‰LˆB��A�qØ" :œÔ4ÑˆC��b˜"Ø�QŠwˆw˜2 š7˜7Ý˜R ™œ¥s¨2¨s¡|¤|�B�Ø36°z´?À3ÀBÀ3ÈÈÈÀ8Ô3L�
˜:Ô& s¨ s¨C¨R¨C xÑ0Ø 1Ñ$�Ý˜u i°Ñ6Ô6ˆŒØð 	yÝŒKÐu ÐuÐu½¸T¼Z×=RÒ=RÑ=TÔ=TÑ9UÔ9UÐuÐuÐuÑvÔvÐvØ8�w�u˜j¨-¨-Ñ8Ô8Ñ:Ô:ð yÀ4Ç=Â=Á?Ä?ð yÝ”ÐwÑxÔxÐxÐxÐxð	yð 	yðyð yð yð yrv   rò   údict[str, torch.Tensor]Ú	src_modelútorch.nn.Moduler¦   Úboolc                ód  ‡‡‡‡— t          |dd¦  «        }t          | dd¦  «        Št          |t          ¦  «        rt          ‰t          ¦  «        sdS t          |¦  «        t          ‰¦  «        c}Šd„ Št	          d„ |‰fD ¦   «         ¦  «        rdS ˆfd„|                     ¦   «         D ¦   «         Št          j        ˆˆˆfd„t          ‰¦  «        D ¦   «         t          j	        ¬¦  «        }t          |dk                         ¦   «         ¦  «        }|dk    s-|‰k    r)t          j        |t          j        ‰¦  «        ¦  «        rdS |dk    }|                      ¦   «         }	ˆfd	„|                      ¦   «         D ¦   «         }
d}|
|                     ¦   «         z  D ]v}||         |	|         }}|j        d
d…         |j        d
d…         k    rŒ3|||                                       |j        ¦  «        ||<   |                     |¦  «         |d
z  }Œw|r|rt+          j        d|› d‰› d�¦  «         |S )aÁ  Remap pretrained classification head rows to current class order by name.

        Copies rows from pretrained cls layers into the current model's state_dict where the destination class name
        matches a source class name (case-insensitive, whitespace-stripped). Useful when fine-tuning across datasets
        with overlapping classes, whether the class counts differ (e.g. Objects365 -> COCO) or match but the class
        order differs. Mutates the destination tensors in-place via state_dict references; matched cls tensors are
        removed from `csd` so the subsequent intersect_dicts skips them.

        Args:
            csd (dict): Pretrained checkpoint state_dict (will be mutated).
            src_model (torch.nn.Module): Pretrained module, used to read `.names` and `.nc`.
            verbose (bool): Log mapping summary.

        Returns:
            (int): Number of cls tensors remapped (counted toward "Transferred" log line).
        ÚnamesNr   c                óh   — t          | ¦  «                             ¦   «                              ¦   «         S ©N©ÚstrÚstripÚlower)Úss    rt   Ú_normz,BaseModel._remap_cls_by_names.<locals>._normw  s"   € Ý�q‘6”6—<’<‘>”>×'Ò'Ñ)Ô)Ð)rv   c              3  ól   K  — | ]/}t          d „ |                     ¦   «         D ¦   «         ¦  «        V — Œ0dS )c              3  ó\   K  — | ]'\  }}t          |¦  «        t          |¦  «        k    V — Œ(d S r  ©r	  ©rƒ   ÚkÚvs      rt   rÔ   z:BaseModel._remap_cls_by_names.<locals>.<genexpr>.<genexpr>{  ó7   è è € Ð:Ð:©¨¨1•3�q‘6”6�S ™VœVÒ#Ð:Ð:Ð:Ð:Ð:Ð:rv   N©ÚallÚitems©rƒ   Úns     rt   rÔ   z0BaseModel._remap_cls_by_names.<locals>.<genexpr>{  óA   è è € Ð[Ð[¸q�sÐ:Ð:°·²±	´	Ð:Ñ:Ô:Ñ:Ô:Ð[Ð[Ð[Ð[Ð[Ð[rv   c                ó.   •— i | ]\  }} ‰|¦  «        |“ŒS r�   r�   )rƒ   r  r  r  s      €rt   ú
<dictcomp>z1BaseModel._remap_cls_by_names.<locals>.<dictcomp>~  s'   ø€ Ð@Ð@Ð@¡d a¨�e�e˜A‘h”h Ð@Ð@Ð@rv   c           	     ót   •— g | ]4}‰                       ‰‰                      |¦  «        ¦  «        d ¦  «        ‘Œ5S r€   ©ré   )rƒ   r  r  Ú
src_lookupÚ	tgt_namess     €€€rt   r†   z1BaseModel._remap_cls_by_names.<locals>.<listcomp>  s=   ø€ Ð_Ð_Ð_ÈA˜JŸNšN¨5¨5°·²¸qÑ1AÔ1AÑ+BÔ+BÀBÑGÔGÐ_Ð_Ð_rv   ©Údtypec                ó  •— h | ]ƒ\  }}t          |t          ¦  «        rid D ]f}t          t          ||d¦  «        ¦  «        D ]E\  }}t          |d         dd¦  «        ‰k    ¯ dD ]"}|› d|› d|› dt	          |¦  «        dz
  › d|› �	’Œ#ŒFŒgŒ„S ))Úcv3Úone2one_cv3r�   r~   Úout_channelsN)ÚweightÚbiasú.r‰   )rl   r)   Ú	enumeraterÓ   rî   )rƒ   ÚnamerŸ   Úattrr”   ÚseqÚpÚtgt_ncs          €rt   ú	<setcomp>z0BaseModel._remap_cls_by_names.<locals>.<setcomp>‰  sÝ   ø€ ð 
ð 
ð 
á��aÝ˜!�VÑ$Ô$ð
ð /ð	
ð 
ð Ý#¥G¨A¨t°RÑ$8Ô$8Ñ9Ô9ð
ð 
ñ
 ��3Ý�s˜2”w °Ñ5Ô5¸Ò?Ð?Ø'ð @Ð?Øð Ð3Ð3�dÐ3Ð3˜QÐ3Ð3¥ S¡¤¨A¡Ð3Ð3°Ð3Ð3ð
 @Ð?Ð?Ð?Ð?Ð?rv   r‰   ú	Remapped ræ   z4 cls head rows from pretrained weights by class name)rÓ   rl   rm   rî   rÕ   r  r–   Útensorr·   Úlongr‘   r»   ÚequalÚarangerë   Únamed_modulesÚkeysrï   Útor"  ÚpoprG   r¸   )rp   rò   r  r¦   Ú	src_namesÚsrc_ncÚidxÚn_matchÚvalidrë   Úcls_keysÚremappedr  Úv_srcÚv_tgtr  r  r   r/  s                  @@@@rt   rì   zBaseModel._remap_cls_by_names`  sj  øøøø€ õ" ˜I w°Ñ5Ô5ˆ	Ý˜D '¨4Ñ0Ô0ˆ	Ý˜9¥dÑ+Ô+ð 	µ
¸9ÅdÑ0KÔ0Kð 	Ø�1Ý˜Y™œ­¨Y©¬ˆˆ�ð	*ð 	*ð 	*õ Ð[Ð[ÀYÐPYÐDZÐ[Ñ[Ô[Ñ[Ô[ð 	Ø�1à@Ð@Ð@Ð@¨i¯oªoÑ.?Ô.?Ð@Ñ@Ô@ˆ
ÝŒlÐ_Ð_Ð_Ð_Ð_Ð_ÕQVÐW]ÑQ^ÔQ^Ð_Ñ_Ô_ÕglÔgqÐrÑrÔrˆÝ�s˜a’x—n’nÑ&Ô&Ñ'Ô'ˆà�aŠ<ˆ<˜F fÒ,Ð,µ´¸SÅ%Ä,ÈvÑBVÔBVÑ1WÔ1WÐ,Ø�1à�q’ˆØ—_’_Ñ&Ô&ˆ
ð
ð 
ð 
ð 
à×-Ò-Ñ/Ô/ð
ñ 
ô 
ˆð ˆØ˜CŸHšH™JœJÑ&ð 	ð 	ˆAØ˜qœ6 :¨a¤=�5ˆEØŒ{˜1˜2˜2Œ %¤+¨a¨b¨b¤/Ò1Ð1ØØ   U¤Ô,×/Ò/°´Ñ<Ô<ˆE�%‰LØ�GŠG�A‰JŒJˆJØ˜‰MˆHˆHØð 	l�xð 	lÝŒKÐj GÐjÐj¨fÐjÐjÐjÑkÔkÐkØˆrv   c                ó¼   — t          | dd¦  «        €|                      ¦   «         | _        |€|                      |d         ¦  «        }|                      ||¦  «        S )ú§Compute loss.

        Args:
            batch (dict): Batch to compute loss on.
            preds (torch.Tensor | list[torch.Tensor], optional): Predictions.
        Ú	criterionNÚimg)rÓ   Úinit_criterionrE  ru   ©rp   ÚbatchÚpredss      rt   rn   zBaseModel.lossž  sW   € õ �4˜ dÑ+Ô+Ð3Ø!×0Ò0Ñ2Ô2ˆDŒNàˆ=Ø—L’L  u¤Ñ.Ô.ˆEØ�~Š~˜e UÑ+Ô+Ð+rv   c                ó    — t          d¦  «        ‚)z0Initialize the loss criterion for the BaseModel.z4compute_loss() needs to be implemented by task heads)ÚNotImplementedErrorrÖ   s    rt   rG  zBaseModel.init_criterion¬  s   € å!Ð"XÑYÔYÐYrv   )FFN)FN)TrÀ   )FTrÀ   ©T©rò   r   r  r  r¦   r  r  )r£   Ú
__module__Ú__qualname__Ú__doc__ru   ro   ry   rx   r’   rÌ   rÄ   r¸   rÛ   rÿ   rì   rn   rG  Ú__classcell__©r¢   s   @rt   rj   rj   y   s3  ø€ € € € € ðð ð00ð 0ð 0ð"5ð 5ð 5ð 5ð ð ð ð ð6%ð %ð %ðIð Ið Ið4%ð %ð %ð %ðN
ð 
ð 
ðQð Qð Qð Qðð ð ð ð ð,yð yð yð yð@<ð <ð <ð <ð <ð|,ð ,ð ,ð ,ðZð Zð Zð Zð Zð Zð Zrv   rj   c                óv  — t          |t          ¦  «        r|nt          |¦  «        | _        | j        d         d         d         dk    r*t	          j        d¦  «         d| j        d         d         d<   || j        d<   |r@|| j        d         k    r/t	          j        d	| j        d         › d
|› �¦  «         || j        d<   t          t          | j        ¦  «        ||¬¦  «        \  | _	        | _
        d„ t          | j        d         ¦  «        D ¦   «         | _        | j                             dd¦  «        | _        dS )z;Initialize common YOLO model attributes from a YAML config.Úbackboner   r§   ÚSilencez�YOLOv9 `Silence` module is deprecated in favor of torch.nn.Identity. Please delete local *.pt file and re-download the latest model checkpoint.znn.IdentityÚchannelsÚncúOverriding model.yaml nc=ú	 with nc=©rý   r¦   c                ó   — i | ]}||› “ŒS r�   r�   ©rƒ   r”   s     rt   r  z*_initialize_yolo_model.<locals>.<dictcomp>À  s   € Ð>Ð>Ð> �1˜�fÐ>Ð>Ð>rv   ÚinplaceTN)rl   rm   Úyaml_model_loadÚyamlrG   r¡   r¸   Úparse_modelr   r�   r•   r·   r  ré   r^  )r�   Úcfgrý   rX  r¦   s        rt   Ú_initialize_yolo_modelrc  ±  s1  € å" 3­Ñ-Ô-ÐG��µ?À3Ñ3GÔ3G€E„JØ„z�*Ô˜aÔ  Ô# yÒ0Ð0ÝŒðYñ	
ô 	
ð 	
ð (5ˆŒ
�:Ô˜qÔ! !Ñ$à€E„JˆzÑØ	ð ˆb�E”J˜tÔ$Ò$Ð$ÝŒÐO°´
¸4Ô0@ÐOÐOÈ2ÐOÐOÑPÔPÐPØˆŒ
�4ÑÝ)­(°5´:Ñ*>Ô*>À2ÈwÐWÑWÔWÑ€E„K�”Ø>Ð>¥e¨E¬J°tÔ,<Ñ&=Ô&=Ð>Ñ>Ô>€E„KØ”J—N’N 9¨dÑ3Ô3€E„M€M€Mrv   c                  óŽ   ‡ — e Zd ZdZdˆ fd„	Zed„ ¦   «         Zej        d„ ¦   «         Zd	„ Zˆ fd
„Z	e
dd„¦   «         Zd„ Zd„ Zˆ xZS )ÚDetectionModelaW  YOLO detection model.

    This class implements the YOLO detection architecture, handling model initialization, forward pass, augmented
    inference, and loss computation for object detection tasks.

    Attributes:
        yaml (dict): Model configuration dictionary.
        model (torch.nn.Sequential): The neural network model.
        save (list): List of layer indices to save outputs from.
        names (dict): Class names dictionary.
        inplace (bool): Whether to use inplace operations.
        end2end (bool): Whether the model uses end-to-end detection.
        stride (torch.Tensor): Model stride values.

    Methods:
        __init__: Initialize the YOLO detection model.
        _predict_augment: Perform augmented inference.
        _descale_pred: De-scale predictions following augmented inference.
        _clip_augmented: Clip YOLO augmented inference tails.
        init_criterion: Initialize the loss criterion.

    Examples:
        Initialize a detection model
        >>> model = DetectionModel("yolo26n.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor)
    úyolo26n.yamlé   NTc                óÀ  •‡ ‡— t          ¦   «                              ¦   «          t          ‰ ||||¦  «         ‰ j        d         }t	          |t
          ¦  «        r°dŠ‰ j        |_        ˆ fd„}‰ j                             ¦   «          d|_        t          j
        ˆfd„ |t          j        d|‰‰¦  «        ¦  «        D ¦   «         ¦  «        |_        |j        ‰ _        ‰ j                             ¦   «          |                     ¦   «          nt          j        dg¦  «        ‰ _        t!          ‰ ¦  «         |r*‰                      ¦   «          t%          j        d¦  «         d	S d	S )
aQ  Initialize the YOLO detection model with the given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r~   é   c                óV   •— ‰                      | ¦  «        }d|v r|d         }|d         S )z_Perform a forward pass through the model, handling different Detect subclass types accordingly.Úone2manyÚfeats)ru   )rq   Úoutputrp   s     €rt   Ú_forwardz)DetectionModel.__init__.<locals>._forwardò  s1   ø€ àŸš a™œ�Ø Ð'Ð'Ø# JÔ/�FØ˜g”Ð&rv   Tc                ó0   •— g | ]}‰|j         d          z  ‘ŒS )éþÿÿÿ)rï   )rƒ   rq   r  s     €rt   r†   z+DetectionModel.__init__.<locals>.<listcomp>û  s#   ø€ Ð$aÐ$aÐ$a¸ Q¨¬°¬¡_Ð$aÐ$aÐ$arv   r‰   é    Ú N)rÚ   Ú__init__rc  r�   rl   r)   r^  ÚevalÚtrainingr–   r2  ÚzerosrÜ   ÚtrainÚ	bias_initÚTensorra   r¸   rG   )	rp   rb  rý   rX  r¦   rŸ   rn  r  r¢   s	   `      @€rt   rs  zDetectionModel.__init__à  sW  øøø€ õ 	‰Œ×ÒÑÔÐÝ˜t S¨"¨b°'Ñ:Ô:Ð:ð ŒJ�rŒNˆÝ�a�Ñ Ô ð 	-ØˆAØœˆAŒIð'ð 'ð 'ð 'ð 'ð ŒJ�OŠOÑÔÐØˆAŒJÝ”|Ð$aÐ$aÐ$aÐ$a¸h¸hÅuÄ{ÐSTÐVXÐZ[Ð]^ÑG_ÔG_Ñ>`Ô>`Ð$aÑ$aÔ$aÑbÔbˆAŒHØœ(ˆDŒKØŒJ×ÒÑÔÐØ�KŠK‰MŒMˆMˆMåœ,¨ tÑ,Ô,ˆDŒKõ 	˜4Ñ Ô Ð Øð 	Ø�IŠI‰KŒKˆKÝŒK˜‰OŒOˆOˆOˆOð	ð 	rv   c                ó:   — t          | j        d         dd¦  «        S )z<Return whether the model uses end-to-end NMS-free detection.r~   Úend2endF)rÓ   r�   rÖ   s    rt   r{  zDetectionModel.end2end  s   € õ �t”z "”~ y°%Ñ8Ô8Ð8rv   c                ón   — t          | j        d         t          ¦  «        r|| j        d         _        dS dS )zESelect the inference head while retaining both branches for training.r~   N)rl   r�   r)   r{  )rp   Úvalues     rt   r{  zDetectionModel.end2end  s:   € õ �d”j ”n¥fÑ-Ô-ð 	+Ø%*ˆDŒJ�rŒNÔ"Ð"Ð"ð	+ð 	+rv   c                óÈ   — | j         d         }|                     ¦   «         D ]?\  }}t          ||¦  «        st          j        d|› d�¦  «         Œ.t          |||¦  «         Œ@dS )zžSet attributes of the model head (last layer).

        Args:
            **kwargs (Any): Arbitrary keyword arguments representing attributes to set.
        r~   zHead has no attribute 'ú'.N)r�   r  rÆ   rG   r¡   Úsetattr)rp   rs   Úheadr  r  s        rt   Úset_head_attrzDetectionModel.set_head_attr  sz   € ð Œz˜"Œ~ˆØ—L’L‘N”Nð 	 ð 	 ‰DˆAˆqÝ˜4 Ñ#Ô#ð Ý”Ð>¸Ð>Ð>Ð>Ñ?Ô?Ð?ØÝ�D˜!˜QÑÔÐÐð		 ð 	 rv   c           	     ó²  •— t          | dd¦  «        s!t          | j        d         ¦  «        t          ur)t	          j        d¦  «         |                      |¦  «        S |j        dd…         }g d¢}g d¢}g }t          ||¦  «        D ]¦\  }}t          |r| 
                    |¦  «        n||t          | j                             ¦   «         ¦  «        ¬	¦  «        }t          ¦   «                              |¦  «        d
         }	|                      |	|||¦  «        }	|                     |	¦  «         Œ§|                      |¦  «        }t'          j        |d¦  «        dfS )a  Perform augmentations on input image x and return augmented inference and train outputs.

        Args:
            x (torch.Tensor): Input image tensor.

        Returns:
            (tuple[torch.Tensor, None]): Augmented inference output and None for train output.
        r{  Fr~   zLModel does not support 'augment=True', reverting to single-scale prediction.rp  N)r‰   g�Âõ(\�ê?gq=
×£på?)Nrg  N)Úgsr   )rÓ   rº   r�   r)   rG   r¡   ry   rï   Úziprf   Úflipr‘   rÜ   rŽ   rÚ   ro   Ú_descale_predr“   Ú_clip_augmentedr–   r›   )rp   rq   Úimg_sizer  r�   r…   ÚsiÚfiÚxiÚyir¢   s             €rt   rx   zDetectionModel._predict_augment   s@  ø€ õ �4˜ EÑ*Ô*ð 	)­d°4´:¸b´>Ñ.BÔ.BÍ&Ð.PÐ.PÝŒNÐiÑjÔjÐjØ×%Ò% aÑ(Ô(Ð(Ø”7˜2˜3˜3”<ˆØˆOˆOˆØˆOˆOˆØˆÝ˜!˜Q‘i”ið 	ð 	‰FˆB�Ý¨Ð2˜1Ÿ6š6 "™:œ:˜:°°B½3¸t¼{¿ºÑ?PÔ?PÑ;QÔ;QÐRÑRÔRˆBÝ‘”—’ Ñ$Ô$ QÔ'ˆBØ×#Ò# B¨¨B°Ñ9Ô9ˆBØ�HŠH�R‰LŒLˆLˆLØ× Ò  Ñ#Ô#ˆÝŒy˜˜BÑÔ Ð%Ð%rv   r‰   c                ó   — | dd…dd…fxx         |z  cc<   |                       ddd| j        |         dz
  f|¦  «        \  }}}}|dk    r|d         |z
  }n|dk    r|d         |z
  }t          j        ||||f|¦  «        S )a³  De-scale predictions following augmented inference (inverse operation).

        Args:
            p (torch.Tensor): Predictions tensor.
            flips (int | None): Flip type (None=none, 2=ud, 3=lr).
            scale (float): Scale factor.
            img_size (tuple): Original image size (height, width).
            dim (int): Dimension to split at.

        Returns:
            (torch.Tensor): De-scaled predictions.
        Né   r‰   r§   r   rg  )Úsplitrï   r–   r›   )	r.  ÚflipsÚscaler‰  r‹   rq   r…   ÚwhÚclss	            rt   r‡  zDetectionModel._descale_pred8  s�   € ð 	
ˆ!ˆ!ˆ!ˆRˆaˆRˆ%ˆˆŒ�EÑˆˆ‰ØŸš  A q¨!¬'°#¬,¸Ñ*:Ð ;¸SÑAÔA‰ˆˆ1ˆb�#Ø�AŠ:ˆ:Ø˜”˜a‘ˆAˆAØ�aŠZˆZØ˜”˜a‘ˆAÝŒy˜!˜Q  C˜¨#Ñ.Ô.Ð.rv   c                óÊ  ‡— | j         d         j        Št          d„ t          ‰¦  «        D ¦   «         ¦  «        }d}|d         j        d         |z  t          d„ t          |¦  «        D ¦   «         ¦  «        z  }|d         dd| …f         |d<   |d         j        d         |z  t          ˆfd„t          |¦  «        D ¦   «         ¦  «        z  }|d         d|d…f         |d<   |S )	zÊClip YOLO augmented inference tails.

        Args:
            y (list[torch.Tensor]): List of detection tensors.

        Returns:
            (list[torch.Tensor]): Clipped detection tensors.
        r~   c              3  ó    K  — | ]	}d |z  V — Œ
dS ©r�  Nr�   ©rƒ   rq   s     rt   rÔ   z1DetectionModel._clip_augmented.<locals>.<genexpr>X  s&   è è € Ð(Ð(˜��1‘Ð(Ð(Ð(Ð(Ð(Ð(rv   r‰   r   c              3  ó    K  — | ]	}d |z  V — Œ
dS r—  r�   r˜  s     rt   rÔ   z1DetectionModel._clip_augmented.<locals>.<genexpr>Z  s&   è è € Ð'?Ð'?°¨¨1©Ð'?Ð'?Ð'?Ð'?Ð'?Ð'?rv   .Nc              3  ó.   •K  — | ]}d ‰dz
  |z
  z  V — ŒdS )r�  r‰   Nr�   )rƒ   rq   Únls     €rt   rÔ   z1DetectionModel._clip_augmented.<locals>.<genexpr>\  s0   øè è € Ð(MÐ(M¸q¨¨r°A©v¸©zÑ):Ð(MÐ(MÐ(MÐ(MÐ(MÐ(Mrv   )r�   r›  r»   r·   rï   )rp   r…   ÚgÚer”   r›  s        @rt   rˆ  zDetectionModel._clip_augmentedN  så   ø€ ð ŒZ˜Œ^ÔˆÝÐ(Ð(�e B™iœiÐ(Ñ(Ô(Ñ(Ô(ˆØˆØˆqŒTŒZ˜Œ^˜qÑ ¥CÐ'?Ð'?µe¸A±h´hÐ'?Ñ'?Ô'?Ñ$?Ô$?Ñ?ˆØ�Œt�C˜˜1˜"˜�HŒ~ˆˆ!‰ØˆrŒUŒ[˜Œ_ Ñ!¥SÐ(MÐ(MÐ(MÐ(MÅEÈ!ÁHÄHÐ(MÑ(MÔ(MÑ%MÔ%MÑMˆØ�"”�c˜1˜2˜2�g”ˆˆ"‰Øˆrv   c                óv   — t          | j        d         dd¦  «        �t          | ¦  «        nt          | ¦  «        S )z5Initialize the loss criterion for the DetectionModel.r~   rÑ   N)rÓ   r�   rU   rY   rÖ   s    rt   rG  zDetectionModel.init_criterion`  s3   € å '¨¬
°2¬¸ÀtÑ LÔ LÐ X�w�t‰}Œ}ˆ}Õ^mÐnrÑ^sÔ^sÐsrv   )rf  rg  NT)r‰   )r£   rO  rP  rQ  rs  Úpropertyr{  Úsetterr‚  rx   Ústaticmethodr‡  rˆ  rG  rR  rS  s   @rt   re  re  Ä  sí   ø€ € € € € ðð ð6&ð &ð &ð &ð &ð &ðP ð9ð 9ñ „Xð9ð „^ð+ð +ñ „^ð+ð
 ð  ð  ð&ð &ð &ð &ð &ð0 ð/ð /ð /ñ „\ð/ð*ð ð ð$tð tð tð tð tð tð trv   re  c                  ó*   ‡ — e Zd ZdZdˆ fd„	Zd„ Zˆ xZS )	ÚOBBModelað  YOLO Oriented Bounding Box (OBB) model.

    This class extends DetectionModel to handle oriented bounding box detection tasks, providing specialized loss
    computation for rotated object detection.

    Methods:
        __init__: Initialize YOLO OBB model.
        init_criterion: Initialize the loss criterion for OBB detection.

    Examples:
        Initialize an OBB model
        >>> model = OBBModel("yolo26n-obb.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor)
    úyolo26n-obb.yamlrg  NTc                óR   •— t          ¦   «                              ||||¬¦  «         dS )aC  Initialize YOLO OBB model with given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        ©rb  rý   rX  r¦   N©rÚ   rs  ©rp   rb  rý   rX  r¦   r¢   s        €rt   rs  zOBBModel.__init__u  ó,   ø€ õ 	‰Œ×Ò˜S R¨B¸ÐÑ@Ô@Ð@Ð@Ð@rv   c                ó‚   — t          | j        d         dd¦  «        �t          | t          ¦  «        nt          | ¦  «        S )z,Initialize the loss criterion for the model.r~   rÑ   N)rÓ   r�   rU   rZ   rÖ   s    rt   rG  zOBBModel.init_criterion€  s9   € å+2°4´:¸b´>À=ÐRVÑ+WÔ+WÐ+c�w�t�YÑ'Ô'Ð'ÕirÐswÑixÔixÐxrv   )r¤  rg  NT©r£   rO  rP  rQ  rs  rG  rR  rS  s   @rt   r£  r£  e  sc   ø€ € € € € ðð ð	Að 	Að 	Að 	Að 	Að 	Aðyð yð yð yð yð yð yrv   r£  c                  ó*   ‡ — e Zd ZdZdˆ fd„	Zd„ Zˆ xZS )	ÚSegmentationModela  YOLO segmentation model.

    This class extends DetectionModel to handle instance segmentation tasks, providing specialized loss computation for
    pixel-level object detection and segmentation.

    Methods:
        __init__: Initialize YOLO segmentation model.
        init_criterion: Initialize the loss criterion for segmentation.

    Examples:
        Initialize a segmentation model
        >>> model = SegmentationModel("yolo26n-seg.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor)
    úyolo26n-seg.yamlrg  NTc                óR   •— t          ¦   «                              ||||¬¦  «         dS )aX  Initialize Ultralytics YOLO segmentation model with given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r¦  Nr§  r¨  s        €rt   rs  zSegmentationModel.__init__•  r©  rv   c                ó‚   — t          | j        d         dd¦  «        �t          | t          ¦  «        nt          | ¦  «        S )z8Initialize the loss criterion for the SegmentationModel.r~   rÑ   N)rÓ   r�   rU   r\   rÖ   s    rt   rG  z SegmentationModel.init_criterion   s?   € õ �t”z "”~ }°dÑ;Ô;ÐGõ �DÕ,Ñ-Ô-Ð-å# DÑ)Ô)ð	
rv   )r®  rg  NTr«  rS  s   @rt   r­  r­  …  s\   ø€ € € € € ðð ð	Að 	Að 	Að 	Að 	Að 	Að
ð 
ð 
ð 
ð 
ð 
ð 
rv   r­  c                  ó4   ‡ — e Zd ZdZd	ˆ fd„	Zd„ Zˆ fd„Zˆ xZS )
ÚSemanticSegmentationModela  YOLO semantic segmentation model.

    This class implements a semantic segmentation model that produces per-pixel class predictions. Unlike
    SegmentationModel (instance segmentation), this does not produce bounding boxes.

    Methods:
        __init__: Initialize the semantic segmentation model.
        init_criterion: Initialize the loss criterion for semantic segmentation.

    Examples:
        Initialize a semantic segmentation model
        >>> model = SemanticSegmentationModel("yolo26n-sem.yaml", ch=3, nc=19)
    úyolo26n-sem.yamlrg  NTc                óH  •‡
— t          ¦   «                              ¦   «          t          | ||||¦  «         | j        d         }t	          |t
          ¦  «        rõd}| j                             ¦   «          d|_        |g}|fd„Š
ˆ
fd„| j        D ¦   «         }	 |                      t          j
        d|||¦  «        ¦  «         |D ]}	|	                     ¦   «          Œn# |D ]}	|	                     ¦   «          Œw xY wt          j        ||d         z  gt          j        ¬¦  «        |_        |j        | _        | j                             ¦   «          nt          j        d	g¦  «        | _        t#          | ¦  «         |r*|                      ¦   «          t'          j        d
¦  «         dS dS )a8  Initialize the YOLO semantic segmentation model.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r~   ri  Tc                ó    — t          |t          j        ¦  «        r1|j        dk    r(t	          |d         |j        d         ¦  «        |d<   d S d S d S )Nr�  r   rp  )rl   r–   ry  Úndimrð   rï   )Ú_mÚ_inpÚoutÚ_hs       rt   Ú_recordz3SemanticSegmentationModel.__init__.<locals>._recordÏ  sO   € Ý˜c¥5¤<Ñ0Ô0ð 6°S´XÀ²]°]Ý  1¤ s¤y°¤}Ñ5Ô5�B�q‘E�E�Eð6ð 6°]°]rv   c                ó:   •— g | ]}|                      ‰¦  «        ‘ŒS r�   )Úregister_forward_hook)rƒ   Úlayerr»  s     €rt   r†   z6SemanticSegmentationModel.__init__.<locals>.<listcomp>Ó  s'   ø€ ÐRÐRÐR¸e�U×0Ò0°Ñ9Ô9ÐRÐRÐRrv   r‰   r   r!  rq  rr  N)rÚ   rs  rc  r�   rl   r?   rt  ru  ru   r–   rv  Úremover2  Úfloat32rÜ   rw  ry  ra   r¸   rG   )rp   rb  rý   rX  r¦   rŸ   r  Úmin_hÚhooksrù   r»  r¢   s             @€rt   rs  z"SemanticSegmentationModel.__init__¸  s®  øø€ õ 	‰Œ×ÒÑÔÐÝ˜t S¨"¨b°'Ñ:Ô:Ð:ð ŒJ�rŒNˆÝ�a�Ñ)Ô)ð 	-ØˆAØŒJ�OŠOÑÔÐØˆAŒJØ�CˆEà*/ð 6ð 6ð 6ð 6ð SÐRÐRÐRÀtÄzÐRÑRÔRˆEðØ—’�Uœ[¨¨B°°1Ñ5Ô5Ñ6Ô6Ð6àð ð �AØ—H’H‘J”J�J�Jðø˜ð ð �AØ—H’H‘J”J�J�Jðøøøå”| Q¨¨q¬¡\ N½%¼-ÐHÑHÔHˆAŒHØœ(ˆDŒKØŒJ×ÒÑÔÐÐåœ,¨ tÑ,Ô,ˆDŒKå˜4Ñ Ô Ð Øð 	Ø�IŠI‰KŒKˆKÝŒK˜‰OŒOˆOˆOˆOð	ð 	s   Â*C ÃC5c                ó    — t          | ¦  «        S )z8Initialize the loss criterion for semantic segmentation.)rW   rÖ   s    rt   rG  z(SemanticSegmentationModel.init_criterionä  s   € å'¨Ñ-Ô-Ð-rv   c                ó¸   •— t          ¦   «                              |¦  «         | j        d         }t          |t          ¦  «        r ||j        ¦  «        |_        | S )z-Apply a function to all tensors in the model.r~   )rÚ   rÛ   r�   rl   r?   rÜ   rà   s      €rt   rÛ   z SemanticSegmentationModel._applyè  sL   ø€ å‰Œ�Š�rÑÔÐØŒJ�rŒNˆÝ�a�Ñ)Ô)ð 	$Ø�r˜!œ(‘|”|ˆAŒHØˆrv   )r³  rg  NT)r£   rO  rP  rQ  rs  rG  rÛ   rR  rS  s   @rt   r²  r²  ©  sp   ø€ € € € € ðð ð*ð *ð *ð *ð *ð *ðX.ð .ð .ðð ð ð ð ð ð ð ð rv   r²  c                  ó*   ‡ — e Zd ZdZd	ˆ fd„	Zd„ Zˆ xZS )
Ú	PoseModela^  YOLO pose model.

    This class extends DetectionModel to handle human pose estimation tasks, providing specialized loss computation for
    keypoint detection and pose estimation.

    Attributes:
        kpt_shape (tuple): Shape of keypoints data (num_keypoints, num_dimensions).

    Methods:
        __init__: Initialize YOLO pose model.
        init_criterion: Initialize the loss criterion for pose estimation.

    Examples:
        Initialize a pose model
        >>> model = PoseModel("yolo26n-pose.yaml", ch=3, nc=1, data_kpt_shape=(17, 3))
        >>> results = model.predict(image_tensor)
    úyolo26n-pose.yamlrg  N©NNTc                óN  •— t          |t          ¦  «        st          |¦  «        }t          |¦  «        rKt	          |¦  «        t	          |d         ¦  «        k    r%t          j        d|d         › d|› �¦  «         ||d<   t          ¦   «                              ||||¬¦  «         dS )al  Initialize Ultralytics YOLO Pose model.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            data_kpt_shape (tuple): Shape of keypoints data.
            verbose (bool): Whether to display model information.
        Ú	kpt_shapez Overriding model.yaml kpt_shape=z with kpt_shape=r¦  N)	rl   rm   r_  rÕ   r²   rG   r¸   rÚ   rs  )rp   rb  rý   rX  Údata_kpt_shaper¦   r¢   s         €rt   rs  zPoseModel.__init__  s«   ø€ õ ˜#�tÑ$Ô$ð 	'Ý! #Ñ&Ô&ˆCÝˆ~ÑÔð 	.¥4¨Ñ#7Ô#7½4ÀÀKÔ@PÑ;QÔ;QÒ#QÐ#QÝŒKÐm¸3¸{Ô;KÐmÐmÐ]kÐmÐmÑnÔnÐnØ-ˆC�ÑÝ‰Œ×Ò˜S R¨B¸ÐÑ@Ô@Ð@Ð@Ð@rv   c                óÌ   — t          | j        d         t          ¦  «        rt          nt          }t          | j        d         dd¦  «        �t          | |¦  «        n
 || ¦  «        S )z0Initialize the loss criterion for the PoseModel.r~   rÑ   N)rl   r�   r5   rV   r[   rÓ   rU   )rp   rn   s     rt   rG  zPoseModel.init_criterion  sY   € å'¨¬
°2¬½Ñ?Ô?ÐO�zˆzÅZˆÝ&-¨d¬j¸¬n¸mÈTÑ&RÔ&RÐ&^�w�t˜TÑ"Ô"Ð"ÐdhÐdhÐimÑdnÔdnÐnrv   )rÇ  rg  NrÈ  Tr«  rS  s   @rt   rÆ  rÆ  ñ  sc   ø€ € € € € ðð ð$Að Að Að Að Að Að"oð oð oð oð oð oð orv   rÆ  c                  ó*   ‡ — e Zd ZdZdˆ fd„	Zd„ Zˆ xZS )	Ú
DepthModelag  YOLO depth estimation model.

    This class extends DetectionModel for monocular depth estimation, using YOLO backbone + FPN with a DPT-style dense
    depth decoder head. Follows the Depth Anything approach adapted to YOLO architecture.

    Examples:
        >>> model = DepthModel("yolo26n-depth.yaml", ch=3)
        >>> results = model(image_tensor)
    úyolo26n-depth.yamlrg  NTc                óR   •— t          ¦   «                              ||||¬¦  «         dS )zInitialize YOLO Depth model.r¦  Nr§  r¨  s        €rt   rs  zDepthModel.__init__&  s*   ø€ å‰Œ×Ò˜S R¨B¸ÐÑ@Ô@Ð@Ð@Ð@rv   c                ó    — t          | ¦  «        S )z$Initialize the depth loss criterion.)rT   rÖ   s    rt   rG  zDepthModel.init_criterion*  s   € å˜4Ñ Ô Ð rv   )rÏ  rg  NTr«  rS  s   @rt   rÎ  rÎ    s\   ø€ € € € € ðð ðAð Að Að Að Að Að!ð !ð !ð !ð !ð !ð !rv   rÎ  c                  óF   ‡ — e Zd ZdZd
ˆ fd„	Zd„ Zed„ ¦   «         Zd	„ Zˆ xZ	S )ÚClassificationModelao  YOLO classification model.

    This class implements the YOLO classification architecture for image classification tasks, providing model
    initialization, configuration, and output reshaping capabilities.

    Attributes:
        yaml (dict): Model configuration dictionary.
        model (torch.nn.Sequential): The neural network model.
        stride (torch.Tensor): Model stride values.
        names (dict): Class names dictionary.

    Methods:
        __init__: Initialize ClassificationModel.
        _from_yaml: Set model configurations and define architecture.
        reshape_outputs: Update model to specified class count.
        init_criterion: Initialize the loss criterion.

    Examples:
        Initialize a classification model
        >>> model = ClassificationModel("yolo26n-cls.yaml", ch=3, nc=1000)
        >>> results = model.predict(image_tensor)
    úyolo26n-cls.yamlrg  NTc                óx   •— t          ¦   «                              ¦   «          |                      ||||¦  «         dS )a\  Initialize ClassificationModel with YAML, channels, number of classes, verbose flag.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        N)rÚ   rs  Ú
_from_yamlr¨  s        €rt   rs  zClassificationModel.__init__G  s9   ø€ õ 	‰Œ×ÒÑÔÐØ�Š˜˜R  WÑ-Ô-Ð-Ð-Ð-rv   c                ó”  — t          |t          ¦  «        r|nt          |¦  «        | _        | j                             d|¦  «        x}| j        d<   |rA|| j        d         k    r0t          j        d| j        d         › d|› �¦  «         || j        d<   n,|s*| j                             dd¦  «        st          d¦  «        ‚t          t          | j        ¦  «        ||¬¦  «        \  | _
        | _        t          j        dg¦  «        | _        d	„ t          | j        d         ¦  «        D ¦   «         | _        |                      ¦   «          dS )
aT  Set Ultralytics YOLO model configurations and define the model architecture.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        rW  rX  rY  rZ  NzFnc not specified. Must specify nc in model.yaml or function arguments.r[  r‰   c                ó   — i | ]}||› “ŒS r�   r�   r]  s     rt   r  z2ClassificationModel._from_yaml.<locals>.<dictcomp>g  s   € Ð@Ð@Ð@ A�a˜A˜Ð@Ð@Ð@rv   )rl   rm   r_  r`  ré   rG   r¸   Ú
ValueErrorra  r   r�   r•   r–   ry  rÜ   r·   r  )rp   rb  rý   rX  r¦   s        rt   rÖ  zClassificationModel._from_yamlS  s5  € õ & c­4Ñ0Ô0ÐJ�C�CµoÀcÑ6JÔ6JˆŒ	ð &*¤Y§]¢]°:¸rÑ%BÔ%BÐBˆˆTŒY�zÑ"Øð 	g�"˜œ	 $œÒ'Ð'ÝŒKÐR°D´I¸d´OÐRÐRÈbÐRÐRÑSÔSÐSØ ˆDŒI�d‰OˆOØð 	g˜DœIŸMšM¨$°Ñ5Ô5ð 	gÝÐeÑfÔfÐfÝ +­H°T´YÑ,?Ô,?ÀBÐPWÐ XÑ XÔ XÑˆŒ
�D”IÝ”l A 3Ñ'Ô'ˆŒØ@Ð@­¨t¬y¸¬Ñ)?Ô)?Ð@Ñ@Ô@ˆŒ
Ø�	Š	‰Œˆˆˆrv   c                óþ  — t          t          | d¦  «        r| j        n|                      ¦   «         ¦  «        d         \  }}t	          |t
          ¦  «        rC|j        j        |k    r1t          j	         
                    |j        j        |¦  «        |_        dS dS t	          |t          j	        j
        ¦  «        rC|j        |k    r6t          | |t          j	         
                    |j        |¦  «        ¦  «         dS dS t	          |t          j	        j        ¦  «        �r]d„ |D ¦   «         }t          j	        j
        |v rƒt          |¦  «        dz
  |ddd…                              t          j	        j
        ¦  «        z
  }||         j        |k    r0t          j	         
                    ||         j        |¦  «        ||<   dS dS t          j	        j        |v rªt          |¦  «        dz
  |ddd…                              t          j	        j        ¦  «        z
  }||         j        |k    r[t          j	                             ||         j        |||         j        ||         j        ||         j        du¬¦  «        ||<   dS dS dS dS )zÅUpdate a TorchVision classification model to class count 'nc' if required.

        Args:
            model (torch.nn.Module): Model to update.
            nc (int): New number of classes.
        r�   r~   c                ó,   — g | ]}t          |¦  «        ‘ŒS r�   )rº   r˜  s     rt   r†   z7ClassificationModel.reshape_outputs.<locals>.<listcomp>z  s   € Ð(Ð(Ð( •T˜!‘W”WÐ(Ð(Ð(rv   r‰   N)r(  )r²   rÆ   r�   Únamed_childrenrl   r#   ÚlinearÚout_featuresr–   r   ÚLinearÚin_featuresr€  Ú
Sequentialrî   ÚindexÚConv2dr&  Úin_channelsÚkernel_sizerÜ   r(  )r�   rX  r+  rŸ   Útypesr”   s         rt   Úreshape_outputsz#ClassificationModel.reshape_outputsj  s<  € õ ¥w¨u°gÑ'>Ô'>ÐI˜œ˜ÀE×YÒYÑ[Ô[Ñ\Ô\Ð]_Ô`‰ˆˆaÝ�a�Ñ"Ô"ð 	ØŒxÔ$¨Ò*Ð*Ý œ8Ÿ?š?¨1¬8Ô+?ÀÑDÔD�”��ð +Ð*å˜�5œ8œ?Ñ+Ô+ð 	ØŒ~ Ò#Ð#Ý˜˜t¥U¤X§_¢_°Q´]ÀBÑ%GÔ%GÑHÔHÐHÐHÐHð $Ð#å˜�5œ8Ô.Ñ/Ô/ñ 	Ø(Ð( aÐ(Ñ(Ô(ˆEÝŒxŒ %Ð'Ð'Ý˜‘J”J ‘N U¨4¨4¨R¨4¤[×%6Ò%6µu´x´Ñ%GÔ%GÑG�Ø�Q”4Ô$¨Ò*Ð*Ý œ8Ÿ?š?¨1¨Q¬4Ô+;¸RÑ@Ô@�A�a‘D�D�Dð +Ð*å”” EÐ)Ð)Ý˜‘J”J ‘N U¨4¨4¨R¨4¤[×%6Ò%6µu´x´Ñ%GÔ%GÑG�Ø�Q”4Ô$¨Ò*Ð*Ý œ8Ÿ?š?Ø˜!œÔ(¨"¨a°¬dÔ.>ÀÀ!ÄÄÐRSÐTUÔRVÔR[ÐcgÐRgð +ñ ô �A�a‘D�D�Dð	ð 	ð *Ð)à*Ð*rv   c                ó   — t          ¦   «         S )z:Initialize the loss criterion for the ClassificationModel.)rX   rÖ   s    rt   rG  z"ClassificationModel.init_criterion†  s   € å#Ñ%Ô%Ð%rv   )rÔ  rg  NT)
r£   rO  rP  rQ  rs  rÖ  r¡  rç  rG  rR  rS  s   @rt   rÓ  rÓ  /  s€   ø€ € € € € ðð ð.
.ð 
.ð 
.ð 
.ð 
.ð 
.ðð ð ð. ðð ñ „\ðð6&ð &ð &ð &ð &ð &ð &rv   rÓ  c                  óD   ‡ — e Zd ZdZdˆ fd„	Zddd„Zd„ Zdd„Zdd„Zˆ xZ	S )ÚRTDETRDetectionModelar  RTDETR (Real-time DEtection and Tracking using Transformers) Detection Model class.

    This class is responsible for constructing the RTDETR architecture, defining loss functions, and facilitating both
    the training and inference processes. RTDETR is an object detection and tracking model that extends from the
    DetectionModel base class.

    Attributes:
        nc (int): Number of classes for detection.
        criterion (RTDETRDetectionLoss): Loss function for training.

    Methods:
        __init__: Initialize the RTDETRDetectionModel.
        init_criterion: Initialize the loss criterion.
        loss: Compute loss for training.
        predict: Perform forward pass through the model.

    Examples:
        Initialize an RTDETR model
        >>> model = RTDETRDetectionModel("rtdetr-l.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor)
    úrtdetr-l.yamlrg  NTc                óR   •— t          ¦   «                              ||||¬¦  «         dS )a.  Initialize the RTDETRDetectionModel.

        Args:
            cfg (str | dict): Configuration file name or path.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Print additional information during initialization.
        r¦  Nr§  r¨  s        €rt   rs  zRTDETRDetectionModel.__init__¢  r©  rv   rò   r   r  r  r¦   r  c                ó"  ‡‡‡— t          |dd¦  «        }t          | dd¦  «        Št          |t          ¦  «        rt          ‰t          ¦  «        sdS t          d„ |‰fD ¦   «         ¦  «        rdS d„ |                     ¦   «         D ¦   «         Št          ‰¦  «        }t          j        ˆˆfd„t          |¦  «        D ¦   «         t          j	        ¬¦  «        }t          |dk                         ¦   «         ¦  «        }|dk    s:t          |¦  «        |k    r)t          j        |t          j        |¦  «        ¦  «        rdS |dk    }|                      ¦   «         Šˆfd„|D ¦   «         }	d}
|	D ]†}||         ‰|         }}|j        |j        k    s |j        d	d…         |j        d	d…         k    rŒC|||                                       |j        ¦  «        ||<   |                     |¦  «         |
d	z  }
Œ‡|r|
rt)          j        d
|› d|› d�¦  «         |
S )aC  Remap RT-DETR decoder cls-head rows by class name.

        Overrides BaseModel's YOLO-specific implementation: RT-DETR's classification tensors live under
        `score_head` and `class_embed` inside `RTDETRDecoder` rather than `Detect.cv3`. All of them are
        row-per-class, including the training-only `denoising_class_embed` embedding, so matched class rows
        transfer even when source and target `nc` differ; any residual shape mismatch is dropped by
        `intersect_dicts`.

        Args:
            csd (dict): Pretrained checkpoint state_dict (will be mutated).
            src_model (torch.nn.Module): Pretrained module, used to read `.names`.
            verbose (bool): Log mapping summary.

        Returns:
            (int): Number of cls tensors remapped (counted toward "Transferred" log line).
        r  Nr   c              3  ól   K  — | ]/}t          d „ |                     ¦   «         D ¦   «         ¦  «        V — Œ0dS )c              3  ó\   K  — | ]'\  }}t          |¦  «        t          |¦  «        k    V — Œ(d S r  r  r  s      rt   rÔ   zERTDETRDetectionModel._remap_cls_by_names.<locals>.<genexpr>.<genexpr>Ã  r  rv   Nr  r  s     rt   rÔ   z;RTDETRDetectionModel._remap_cls_by_names.<locals>.<genexpr>Ã  r  rv   c                ó|   — i | ]9\  }}t          |¦  «                             ¦   «                              ¦   «         |“Œ:S r�   r  r  s      rt   r  z<RTDETRDetectionModel._remap_cls_by_names.<locals>.<dictcomp>Æ  s8   € ÐNÐNÐN±D°A°q•c˜!‘f”f—l’l‘n”n×*Ò*Ñ,Ô,¨aÐNÐNÐNrv   c                óª   •— g | ]O}‰                      t          ‰|         ¦  «                             ¦   «                              ¦   «         d ¦  «        ‘ŒPS r€   )ré   r	  r
  r  )rƒ   r  r  r   s     €€rt   r†   z<RTDETRDetectionModel._remap_cls_by_names.<locals>.<listcomp>É  sK   ø€ ÐZÐZÐZÀqˆZ�^Š^�C 	¨!¤Ñ-Ô-×3Ò3Ñ5Ô5×;Ò;Ñ=Ô=¸rÑBÔBÐZÐZÐZrv   r!  c                ó,   •— h | ]}d |v sd|v ¯
|‰v ¯|’ŒS )Ú
score_headÚclass_embedr�   )rƒ   r  rë   s     €rt   r0  z;RTDETRDetectionModel._remap_cls_by_names.<locals>.<setcomp>Ò  s9   ø€ ÐdÐdÐd˜! |°qÐ'8Ð'8¸MÈQÐ<NÐ<NÐTUÐYcÐTcÐTc�AÐTcÐTcÐTcrv   r‰   r1  ræ   z< decoder cls head rows from pretrained weights by class name)rÓ   rl   rm   rÕ   r  rî   r–   r2  r·   r3  r‘   r»   r4  r5  rë   r¶  rï   r8  r"  r9  rG   r¸   )rp   rò   r  r¦   r:  r/  r<  r=  r>  r?  r@  r  rA  rB  r  rë   r   s                 @@@rt   rì   z(RTDETRDetectionModel._remap_cls_by_names­  s3  øøø€ õ" ˜I w°Ñ5Ô5ˆ	Ý˜D '¨4Ñ0Ô0ˆ	Ý˜9¥dÑ+Ô+ð 	µ
¸9ÅdÑ0KÔ0Kð 	Ø�1åÐ[Ð[ÀYÐPYÐDZÐ[Ñ[Ô[Ñ[Ô[ð 	Ø�1àNÐN¸I¿OºOÑ<MÔ<MÐNÑNÔNˆ
Ý�Y‘”ˆÝŒlØZÐZÐZÐZÐZÍEÐRXÉMÌMÐZÑZÔZÕbgÔblð
ñ 
ô 
ˆõ �s˜a’x—n’nÑ&Ô&Ñ'Ô'ˆà�aŠ<ˆ<�C 	™NœN¨fÒ4Ð4½¼ÀSÍ%Ì,ÐW]ÑJ^ÔJ^Ñ9_Ô9_Ð4Ø�1à�q’ˆØ—_’_Ñ&Ô&ˆ
ØdÐdÐdÐd˜sÐdÑdÔdˆØˆØð 	ð 	ˆAØ˜qœ6 :¨a¤=�5ˆEØŒz˜UœZÒ'Ð'¨5¬;°q°r°r¬?¸e¼kÈ!È"È"¼oÒ+MÐ+MØØ   U¤Ô,×/Ò/°´Ñ<Ô<ˆE�%‰LØ�GŠG�A‰JŒJˆJØ˜‰MˆHˆHØð 	t�xð 	tÝŒKÐr GÐrÐr¨fÐrÐrÐrÑsÔsÐsØˆrv   c                ó2   — ddl m}  || j        d¬¦  «        S )z;Initialize the loss criterion for the RTDETRDetectionModel.r   )ÚRTDETRDetectionLossT)rX  Úuse_vfl)Úultralytics.models.utils.lossrö  rX  )rp   rö  s     rt   rG  z#RTDETRDetectionModel.init_criterionß  s,   € àEÐEÐEÐEÐEÐEà"Ð" d¤g°tÐ<Ñ<Ô<Ð<rv   c                ó¤  ‡— t          | d¦  «        s|                      ¦   «         | _        |d         }|j        d         }|d         Šˆfd„t	          |¦  «        D ¦   «         }|d                              |j        t          j        ¬¦  «         	                    d¦  «        |d	                              |j        ¬
¦  «        ‰                     |j        t          j        ¬¦  «         	                    d¦  «        |dœ}|€|  
                    ||¬¦  «        }| j        r|n|d         \  }}}	}
}|€d\  }}n@t          j        ||d         d¬¦  «        \  }}t          j        ||d         d¬¦  «        \  }}t          j        |	                     d¦  «        |g¦  «        }t          j        |
                     d¦  «        |g¦  «        }|                      ||f||||¬¦  «        }t          |                     ¦   «         ¦  «        |d                              ¦   «         |d                              ¦   «         |d                              ¦   «         dœfS )a@  Compute the loss for the given batch of data.

        Args:
            batch (dict): Dictionary containing image and label data.
            preds (tuple, optional): Precomputed model predictions.

        Returns:
            (torch.Tensor): Total loss value.
            (dict): Main three losses in a dict.
        rE  rF  r   Ú	batch_idxc                ód   •— g | ],}‰|k                          ¦   «                              ¦   «         ‘Œ-S r�   )r»   Úitem)rƒ   r”   rú  s     €rt   r†   z-RTDETRDetectionModel.loss.<locals>.<listcomp>÷  s5   ø€ ÐFÐFÐF°q�i 1’n×)Ò)Ñ+Ô+×0Ò0Ñ2Ô2ÐFÐFÐFrv   r”  r!  r~   Úbboxes©r¶   )r”  rý  rú  Ú	gt_groupsN)rI  r‰   rÈ  Údn_num_splitr§   rŠ   )Ú	dn_bboxesÚ	dn_scoresÚdn_metaÚ	loss_giouÚ
loss_classÚ	loss_bbox)Ú	giou_lossÚcls_lossÚl1_loss)rÆ   rG  rE  rï   r·   r8  r¶   r–   r3  Úviewro   ru  r�  r›   Ú	unsqueezer»   ÚvaluesÚdetach)rp   rI  rJ  rF  Úbsrÿ  ÚtargetsÚ
dec_bboxesÚ
dec_scoresÚ
enc_bboxesÚ
enc_scoresr  r  r  rn   rú  s                  @rt   rn   zRTDETRDetectionModel.losså  sA  ø€ õ �t˜[Ñ)Ô)ð 	3Ø!×0Ò0Ñ2Ô2ˆDŒNà�EŒlˆàŒY�qŒ\ˆØ˜+Ô&ˆ	ØFÐFÐFÐF½EÀ"¹I¼IÐFÑFÔFˆ	à˜”<—?’? 3¤:µU´Z�?Ñ@Ô@×EÒEÀbÑIÔIØ˜H”o×(Ò(°´
Ð(Ñ;Ô;Ø"Ÿš c¤j½¼
˜ÑCÔC×HÒHÈÑLÔLØ"ð	
ð 
ˆð ˆ=Ø—L’L ¨G�LÑ4Ô4ˆEØKOÌ=ÐBfÀ%À%Ð^cÐdeÔ^fÑ?ˆ
�J 
¨J¸Øˆ?Ø#-Ñ ˆI�y�yå$)¤K°
¸GÀNÔ<SÐYZÐ$[Ñ$[Ô$[Ñ!ˆI�zÝ$)¤K°
¸GÀNÔ<SÐYZÐ$[Ñ$[Ô$[Ñ!ˆI�zå”Y 
× 4Ò 4°QÑ 7Ô 7¸ÐDÑEÔEˆ
Ý”Y 
× 4Ò 4°QÑ 7Ô 7¸ÐDÑEÔEˆ
à�~Š~Ø˜Ð$ g¸ÈiÐahð ñ 
ô 
ˆõ �4—;’;‘=”=Ñ!Ô!Ø˜kÔ*×1Ò1Ñ3Ô3Ø˜\Ô*×1Ò1Ñ3Ô3Ø˜KÔ(×/Ò/Ñ1Ô1ð$
ð $
ð 
ð 	
rv   Fc                ó2  ‡‡— g g g cŠ}}|rt          |¦  «        ndh}t          |¦  «        }| j        dd…         D �]/}	|	j        dk    r;t	          |	j        t
          ¦  «        r‰|	j                 nˆˆfd„|	j        D ¦   «         Š|r|                      |	‰|¦  «          |	‰¦  «        Š‰                     |	j        | j	        v r‰nd¦  «         |	j        |v r”|                     t          j        j                             ‰d¦  «                             d¦  «                             d¦  «        ¦  «         |	j        |k    r+t          j        t          j        |d¦  «        d¬¦  «        c S �Œ1| j        d         }
 |
ˆfd„|
j        D ¦   «         |¦  «        Š‰S )	aö  Perform a forward pass through the model.

        Args:
            x (torch.Tensor): The input tensor.
            profile (bool): If True, profile the computation time for each layer.
            batch (dict, optional): Ground truth data for evaluation.
            augment (bool): If True, perform data augmentation during inference.
            embed (list, optional): A list of layer indices to return embeddings from.

        Returns:
            (torch.Tensor): Model's output tensor.
        r~   Nc                ó0   •— g | ]}|d k    r‰n‰|         ‘ŒS r€   r�   r‚   s     €€rt   r†   z0RTDETRDetectionModel.predict.<locals>.<listcomp>'  r‡   rv   rˆ   r‰   r   rŠ   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   )rƒ   r„   r…   s     €rt   r†   z0RTDETRDetectionModel.predict.<locals>.<listcomp>1  s   ø€ Ð'Ð'Ð'˜1�!�A”$Ð'Ð'Ð'rv   rŒ   )rp   rq   rz   rI  r{   r|   rœ   r�   rž   rŸ   r�  r…   s    `         @rt   ro   zRTDETRDetectionModel.predict  sª  øø€ ð   BÐˆˆ2ˆzØ$)Ð3•	˜%Ñ Ô Ð °¨tˆÝ�e‘*”*ˆØ”˜C˜R˜C”ð 
	Iñ 
	IˆAØŒs�bŠyˆyÝ(¨¬­cÑ2Ô2Ð]�A�a”c”F�FÐ8]Ð8]Ð8]Ð8]Ð8]ÐYZÔY\Ð8]Ñ8]Ô8]�Øð 2Ø×'Ò'¨¨1¨bÑ1Ô1Ð1Ø��!‘”ˆAØ�HŠH˜!œ# ¤Ð*Ð*�Q�Q°Ñ5Ô5Ð5ØŒs�eˆ|ˆ|Ø×!Ò!¥%¤(Ô"5×"IÒ"IÈ!ÈVÑ"TÔ"T×"\Ò"\Ð]_Ñ"`Ô"`×"hÒ"hÐikÑ"lÔ"lÑmÔmÐmØ”3˜'’>�>Ý œ<­¬	°*¸aÑ(@Ô(@ÀaÐHÑHÔHÐHÐHÐHùØŒz˜"Œ~ˆØˆDÐ'Ð'Ð'Ð' ¤Ð'Ñ'Ô'¨Ñ/Ô/ˆØˆrv   )rë  rg  NTrM  rN  r  ©FNFN)
r£   rO  rP  rQ  rs  rì   rG  rn   ro   rR  rS  s   @rt   rê  rê  ‹  sŸ   ø€ € € € € ðð ð,	Að 	Að 	Að 	Að 	Að 	Að0ð 0ð 0ð 0ð 0ðd=ð =ð =ð.
ð .
ð .
ð .
ð`ð ð ð ð ð ð ð rv   rê  c                  óD   ‡ — e Zd ZdZdˆ fd„	Zdd„Zdd	„Zdd„Zdd„Zˆ xZ	S )Ú
WorldModelap  YOLOv8 World Model.

    This class implements the YOLOv8 World model for open-vocabulary object detection, supporting text-based class
    specification and CLIP model integration for zero-shot detection capabilities.

    Attributes:
        txt_feats (torch.Tensor): Text feature embeddings for classes.
        clip_model (torch.nn.Module): CLIP model for text encoding.

    Methods:
        __init__: Initialize YOLOv8 world model.
        set_classes: Set classes for offline inference.
        get_text_pe: Get text positional embeddings.
        predict: Perform forward pass with text features.
        loss: Compute loss with text features.

    Examples:
        Initialize a world model
        >>> model = WorldModel("yolov8s-world.yaml", ch=3, nc=80)
        >>> model.set_classes(["person", "car", "bicycle"])
        >>> results = model.predict(image_tensor)
    úyolov8s-world.yamlrg  NTc                óš   •— t          j        d|pdd¦  «        | _        d| _        t	          ¦   «                              ||||¬¦  «         dS )aG  Initialize YOLOv8 world model with given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r‰   éP   é   Nr¦  )r–   ÚrandnÚ	txt_featsÚ
clip_modelrÚ   rs  r¨  s        €rt   rs  zWorldModel.__init__M  sL   ø€ õ œ Q¨¨¨b°#Ñ6Ô6ˆŒØˆŒÝ‰Œ×Ò˜S R¨B¸ÐÑ@Ô@Ð@Ð@Ð@rv   r  c                ó~   — |                       |||¬¦  «        | _        t          |¦  «        | j        d         _        dS )a$  Set classes in advance so that model could do offline-inference without clip model.

        Args:
            text (list[str]): List of class names.
            batch (int): Batch size for processing text tokens.
            cache_clip_model (bool): Whether to cache the CLIP model.
        )rI  Úcache_clip_modelr~   N)Úget_text_per  rî   r�   rX  )rp   ÚtextrI  r"  s       rt   Úset_classeszWorldModel.set_classesZ  s;   € ð ×)Ò)¨$°eÐN^Ð)Ñ_Ô_ˆŒÝ ™IœIˆŒ
�2ŒÔÐÐrv   c                ó  ‡— ddl m} t          | j                             ¦   «         ¦  «        j        }t          | dd¦  «        s|r |d|¬¦  «        | _        |r| j        n |d|¬¦  «        Š‰                     |¦  «        }ˆfd„| 	                    |¦  «        D ¦   «         }t          |¦  «        dk    r|d         nt          j        |d¬	¦  «        }|                     d
t          |¦  «        |j        d
         ¦  «        S )aO  Get text positional embeddings using the CLIP model.

        Args:
            text (list[str]): List of class names.
            batch (int): Batch size for processing text tokens.
            cache_clip_model (bool): Whether to cache the CLIP model.

        Returns:
            (torch.Tensor): Text positional embeddings.
        r   ©Úbuild_text_modelr   Nzclip:ViT-B/32rþ  c                ó^   •— g | ])}‰                      |¦  «                             ¦   «         ‘Œ*S r�   ©Úencode_textr  ©rƒ   Útokenr�   s     €rt   r†   z*WorldModel.get_text_pe.<locals>.<listcomp>x  ó3   ø€ Ð\Ð\Ð\¸5�U×&Ò& uÑ-Ô-×4Ò4Ñ6Ô6Ð\Ð\Ð\rv   r‰   rŠ   r~   )Úultralytics.nn.text_modelr(  r´   r�   rµ   r¶   rÓ   r   Útokenizer�  rî   r–   r›   Úreshaperï   )	rp   r$  rI  r"  r(  r¶   Ú
text_tokenr  r�   s	           @rt   r#  zWorldModel.get_text_pee  s  ø€ ð 	?Ð>Ð>Ð>Ð>Ð>å�d”j×+Ò+Ñ-Ô-Ñ.Ô.Ô5ˆÝ�t˜\¨4Ñ0Ô0ð 	OÐ5Eð 	Oà.Ð.¨ÀvÐNÑNÔNˆDŒOØ#3Ði�”�Ð9IÐ9IÈ/ÐbhÐ9iÑ9iÔ9iˆØ—^’^ DÑ)Ô)ˆ
Ø\Ð\Ð\Ð\ÀJ×DTÒDTÐUZÑD[ÔD[Ð\Ñ\Ô\ˆ	Ý$'¨	¡N¤N°aÒ$7Ð$7�I˜a”L�L½U¼YÀyÐVWÐ=XÑ=XÔ=Xˆ	Ø× Ò  ¥S¨¡Y¤Y°	´ÀÔ0CÑDÔDÐDrv   Fc                óž  ‡‡— |€| j         n|                     ‰¦  «        }|j        d         ‰j        d         k    s| j        d         j        r"|                     ‰j        d         dd¦  «        }|                     ¦   «         }g g g cŠ}}|rt          |¦  «        ndh}t          |¦  «        }	| j        D �]•}
|
j	        dk    r;t          |
j	        t          ¦  «        r‰|
j	                 nˆˆfd„|
j	        D ¦   «         Š|r|                      |
‰|¦  «         t          |
t          ¦  «        r |
‰|¦  «        ŠnOt          |
t          ¦  «        r |
‰|¦  «        Šn-t          |
t          ¦  «        r |
‰|¦  «        }n |
‰¦  «        Š‰                     |
j        | j        v r‰nd¦  «         |
j        |v r”|                     t&          j        j                             ‰d¦  «                             d¦  «                             d¦  «        ¦  «         |
j        |	k    r+t'          j        t'          j        |d¦  «        d¬¦  «        c S �Œ—‰S )a	  Perform a forward pass through the model.

        Args:
            x (torch.Tensor): The input tensor.
            profile (bool): If True, profile the computation time for each layer.
            txt_feats (torch.Tensor, optional): The text features, use it if it's given.
            augment (bool): If True, perform data augmentation during inference.
            embed (list, optional): A list of layer indices to return embeddings from.

        Returns:
            (torch.Tensor): Model's output tensor.
        Nr   r~   c                ó0   •— g | ]}|d k    r‰n‰|         ‘ŒS r€   r�   r‚   s     €€rt   r†   z&WorldModel.predict.<locals>.<listcomp>’  r‡   rv   rˆ   r‰   rŠ   )r  Útype_asrï   r�   ÚexportÚexpandÚcloner�   rŽ   r�   rl   r‘   r’   r   rA   r1   r“   r”   r•   r–   r   r—   r˜   r™   rš   r›   )rp   rq   rz   r  r{   r|   Úori_txt_featsrœ   r�   rž   rŸ   r…   s    `         @rt   ro   zWorldModel.predict|  sL  øø€ ð (1Ð'8�T”^�^¸i×PÒPÐQRÑSÔSˆ	ØŒ?˜1Ô ¤¨¤Ò+Ð+¨t¬z¸"¬~Ô/DÐ+Ø!×(Ò(¨¬°¬°R¸Ñ<Ô<ˆIØ!ŸšÑ)Ô)ˆØ  BÐˆˆ2ˆzØ$)Ð3•	˜%Ñ Ô Ð °¨tˆÝ�e‘*”*ˆØ”ð 	Iñ 	IˆAØŒs�bŠyˆyÝ(¨¬­cÑ2Ô2Ð]�A�a”c”F�FÐ8]Ð8]Ð8]Ð8]Ð8]ÐYZÔY\Ð8]Ñ8]Ô8]�Øð 2Ø×'Ò'¨¨1¨bÑ1Ô1Ð1Ý˜!�WÑ%Ô%ð Ø�A�a˜‘O”O��Ý˜A�{Ñ+Ô+ð Ø�A�a˜Ñ'Ô'��Ý˜AÕ/Ñ0Ô0ð Ø˜A˜a ™OœO�	�	à�A�a‘D”D�à�HŠH˜!œ# ¤Ð*Ð*�Q�Q°Ñ5Ô5Ð5ØŒs�eˆ|ˆ|Ø×!Ò!¥%¤(Ô"5×"IÒ"IÈ!ÈVÑ"TÔ"T×"\Ò"\Ð]_Ñ"`Ô"`×"hÒ"hÐikÑ"lÔ"lÑmÔmÐmØ”3˜'’>�>Ý œ<­¬	°*¸aÑ(@Ô(@ÀaÐHÑHÔHÐHÐHÐHùØˆrv   c                óÊ   — t          | d¦  «        s|                      ¦   «         | _        |€#|                      |d         |d         ¬¦  «        }|                      ||¦  «        S )rD  rE  NrF  r  )r  )rÆ   rG  rE  ru   rH  s      rt   rn   zWorldModel.loss¥  s`   € õ �t˜[Ñ)Ô)ð 	3Ø!×0Ò0Ñ2Ô2ˆDŒNàˆ=Ø—L’L  u¤¸¸{Ô9K�LÑLÔLˆEØ�~Š~˜e UÑ+Ô+Ð+rv   )r  rg  NT)r  Tr  r  )
r£   rO  rP  rQ  rs  r%  r#  ro   rn   rR  rS  s   @rt   r  r  5  s¢   ø€ € € € € ðð ð.Að Að Að Að Að Að	&ð 	&ð 	&ð 	&ðEð Eð Eð Eð.'ð 'ð 'ð 'ðR,ð ,ð ,ð ,ð ,ð ,ð ,ð ,rv   r  c                  óœ   ‡ — e Zd ZdZdˆ fd„	Z e¦   «         dd	„¦   «         Z e¦   «         d
„ ¦   «         Zdd„Zd„ Z	d„ Z
d„ Zdd„Zdd„Zˆ xZS )Ú
YOLOEModelaÝ  YOLOE detection model.

    This class implements the YOLOE architecture for efficient object detection with text and visual prompts, supporting
    both prompt-based and prompt-free inference modes.

    Attributes:
        pe (torch.Tensor): Prompt embeddings for classes.
        clip_model (torch.nn.Module): CLIP model for text encoding.

    Methods:
        __init__: Initialize YOLOE model.
        get_text_pe: Get text positional embeddings.
        get_visual_pe: Get visual embeddings.
        set_vocab: Set vocabulary for prompt-free model.
        get_vocab: Get fused vocabulary layer.
        set_classes: Set classes for offline inference.
        get_cls_pe: Get class positional embeddings.
        predict: Perform forward pass with prompts.
        loss: Compute loss with prompts.

    Examples:
        Initialize a YOLOE model
        >>> model = YOLOEModel("yoloe-v8s.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor, tpe=text_embeddings)
    úyoloe-v8s.yamlrg  NTc                ó’   •— t          ¦   «                              ||||¬¦  «         | j                             dd¦  «        | _        dS )a@  Initialize YOLOE model with given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r¦  Ú
text_modelúmobileclip:bltN)rÚ   rs  r`  ré   r?  r¨  s        €rt   rs  zYOLOEModel.__init__Ï  sB   ø€ õ 	‰Œ×Ò˜S R¨B¸ÐÑ@Ô@Ð@Øœ)Ÿ-š-¨Ð6FÑGÔGˆŒˆˆrv   r  Fc                ó*  ‡— ddl m} t          |¦  «        sJ d|› �¦   «         ‚t          | j                             ¦   «         ¦  «        }|j        }t          | dd¦  «        s#|r! |t          | dd¦  «        |¬¦  «        | _        |r| j        n |t          | dd¦  «        |¬¦  «        Š‰ 	                    |¦  «        }ˆfd	„| 
                    |¦  «        D ¦   «         }	t          |	¦  «        d
k    r|	d         nt          j        |	d¬¦  «        }	|	                     dt          |¦  «        |	j        d         ¦  «                             |j        ¦  «        }	|r|	S | j        d         }
t#          |
t$          ¦  «        sJ ‚|
                     |	¦  «        S )aÕ  Get text positional embeddings using the CLIP model.

        Args:
            text (list[str]): List of class names.
            batch (int): Batch size for processing text tokens.
            cache_clip_model (bool): Whether to cache the CLIP model.
            without_reprta (bool): Whether to return text embeddings without reprta module processing.

        Returns:
            (torch.Tensor): Text positional embeddings in the model's parameter dtype.
        r   r'  z*Expected at least one class name, but got r   Nr?  r@  rþ  c                ó^   •— g | ])}‰                      |¦  «                             ¦   «         ‘Œ*S r�   r*  r,  s     €rt   r†   z*YOLOEModel.get_text_pe.<locals>.<listcomp>÷  r.  rv   r‰   rŠ   r~   )r/  r(  rî   r´   r�   rµ   r¶   rÓ   r   r0  r�  r–   r›   r1  rï   r8  r"  rl   rB   Úget_tpe)rp   r$  rI  r"  Úwithout_reprtar(  Úparamr¶   r2  r  r�  r�   s              @rt   r#  zYOLOEModel.get_text_peÛ  s²  ø€ ð 	?Ð>Ð>Ð>Ð>Ð>å�4‰yŒyÐMÐMÐMÀtÐMÐMÑMÔMˆyÝ�T”Z×*Ò*Ñ,Ô,Ñ-Ô-ˆØ”ˆÝ�t˜\¨4Ñ0Ô0ð 	mÐ5Eð 	mà.Ð.­w°t¸\ÐK[Ñ/\Ô/\ÐekÐlÑlÔlˆDŒOð  ð`ˆDŒOˆOà!Ð!¥'¨$°Ð>NÑ"OÔ"OÐX^Ð_Ñ_Ô_ð 	ð
 —^’^ DÑ)Ô)ˆ
Ø\Ð\Ð\Ð\ÀJ×DTÒDTÐUZÑD[ÔD[Ð\Ñ\Ô\ˆ	Ý$'¨	¡N¤N°aÒ$7Ð$7�I˜a”L�L½U¼YÀyÐVWÐ=XÑ=XÔ=Xˆ	Ø×%Ò% b­#¨d©)¬)°Y´_ÀRÔ5HÑIÔI×LÒLÈUÌ[ÑYÔYˆ	Øð 	ØÐàŒz˜"Œ~ˆÝ˜$¥Ñ,Ô,Ð,Ð,Ð,Ø�|Š|˜IÑ&Ô&Ð&rv   c                ó   —  | ||d¬¦  «        S )zíGet visual positional embeddings.

        Args:
            img (torch.Tensor): Input image tensor.
            visual (torch.Tensor): Visual features.

        Returns:
            (torch.Tensor): Visual positional embeddings.
        T)ÚvpeÚ
return_vper�   )rp   rF  Úvisuals      rt   Úget_visual_pezYOLOEModel.get_visual_pe  s   € ð ˆt�C˜V°Ð5Ñ5Ô5Ð5rv   c           
     ó   ‡— | j         rJ ‚| j        d         Št          ‰t          ¦  «        sJ ‚t	          |¦  «        }t          j        ¦   «         5   | t          |                      ¦   «         ¦  «         	                    dd| j
        d         | j
        d         ¦  «        ¦  «         ddd¦  «         n# 1 swxY w Y   |€5‰j        rdnd}d|t          ‰|› d	�¦  «        t          ‰|› d
�¦  «        fi}n|‰j        ‰j        f|‰j        ‰j        fdœ}t#          ˆfd„|                     ¦   «         D ¦   «         ¦  «        sJ d‰j        › d�¦   «         ‚|                     ¦   «         D ]¨\  }\  }}}	d„ t+          t-          ||	|¦  «        ¦  «        D ¦   «         }
t/          ‰|t1          j        |
¦  «        ¦  «         t-          ||	¦  «        D ]C\  }}t          |t0          j        ¦  «        sJ ‚t          |t0          j        ¦  «        sJ ‚|d= |d= ŒDŒ©|€‰                     ¦   «          t9          |¦  «        | j        d         _        || _        dS )a+  Set vocabulary for the prompt-free model.

        Args:
            vocab (nn.ModuleList): One-to-many vocabulary items returned by ``get_vocab`` for ``names``.
            names (list[str]): List of class names.
            one2one_vocab (nn.ModuleList | None): One-to-one vocabulary items. When provided, both the one-to-many and
                one-to-one prompt-free heads are built, so the ``nms`` argument keeps selecting between them at
                inference. When omitted, only the branch selected by ``end2end`` is built, as before.
        r~   r‰   rg  rÃ   NÚone2one_rr  ÚlrpcrÒ   r$  )rM  Úone2one_lrpcc              3  óP   •K  — | ] \  }}}t          |¦  «        ‰j        k    V — Œ!d S r  )rî   r›  )rƒ   r  r¿   r�  s      €rt   rÔ   z'YOLOEModel.set_vocab.<locals>.<genexpr>)  s6   øè è € ÐFÐF©¨¨A¨q•3�q‘6”6˜TœWÒ$ÐFÐFÐFÐFÐFÐFrv   zEach vocabulary needs z items.c              3  ój   K  — | ].\  }\  }}}t          ||d          |d          |dk    ¬¦  «        V — Œ/dS )r~   r§   )ÚenabledN)r3   )rƒ   r”   r”  ÚpfÚlocs        rt   rÔ   z'YOLOEModel.set_vocab.<locals>.<genexpr>+  sL   è è € ÐxÐxÑGXÀqÉ.È3ÐPRÐTW•H˜S " R¤&¨#¨b¬'¸1Àº6ÐBÑBÔBÐxÐxÐxÐxÐxÐxrv   )ru  r�   rl   rB   r   r–   Úno_gradr´   rµ   Ú	new_emptyrr   r{  rÓ   rÒ   r$  rÑ   r%  r  r  r›  r  r*  r…  r€  r   Ú
ModuleListrá  rÌ   rî   rX  r  )rp   Úvocabr  Úone2one_vocabÚe2eÚbranchesr+  r  rÒ   r$  rM  Úloc_headÚcls_headr�  s                @rt   Ú	set_vocabzYOLOEModel.set_vocab  sÁ  ø€ ð ”=Ð Ð Ð ØŒz˜"Œ~ˆÝ˜$¥Ñ,Ô,Ð,Ð,Ð,Ý! %Ñ(Ô(ˆåŒ]‰_Œ_ð 	bð 	bØˆD•�d—o’oÑ'Ô'Ñ(Ô(×2Ò2°1°a¸¼À7Ô9KÈTÌYÐW^ÔM_Ñ`Ô`ÑaÔaÐað	bð 	bð 	bñ 	bô 	bð 	bð 	bð 	bð 	bð 	bð 	bøøøð 	bð 	bð 	bð 	bð Ð Ø $¤Ð4�*�*°"ˆCØ ­°¸°k°k°kÑ(BÔ(BÅGÈDÐUXÐR]ÐR]ÐR]ÑD^ÔD^Ð _Ð`ˆHˆHð  ¤¨$¬(Ð3Ø!.°Ô0@À$ÔBRÐ Sðð ˆHõ ÐFÐFÐFÐF°H·O²OÑ4EÔ4EÐFÑFÔFÑFÔFÐqÐqÐHqÐaeÔahÐHqÐHqÐHqÑqÔqÐFØ#+§>¢>Ñ#3Ô#3ð 	!ð 	!ÑˆD‘-�1�c˜3ØxÐxÕ\eÕfiÐjkÐmpÐruÑfvÔfvÑ\wÔ\wÐxÑxÔxˆDÝ�D˜$¥¤¨dÑ 3Ô 3Ñ4Ô4Ð4Ý&)¨#¨s¡m¤mð !ð !Ñ"�˜(Ý! (­B¬MÑ:Ô:Ð:Ð:Ð:Ý! (­B¬MÑ:Ô:Ð:Ð:Ð:Ø˜R�LØ˜R�L�Lð	!ð
 Ð Ø�IŠI‰KŒKˆKÝ ™JœJˆŒ
�2ŒÔØˆŒ
ˆ
ˆ
s   ÁAB4Â4B8Â;B8c                óœ  — | j         rJ ‚| j        d         }t          |t          ¦  «        sJ ‚|j        rJ ‚t          t          |¦  «                             ¦   «         ¦  «        }|                      |¦  «        }|  	                    ||¦  «         t          | j                             ¦   «         ¦  «        j        }|                     | j                             |¦  «        ¦  «         |j        r|j        n|j        }t'          j        ¦   «         }|D ]9}t          |t&          j        ¦  «        sJ ‚|                     |d         ¦  «         Œ:|S )zÁGet fused vocabulary layer from the model.

        Args:
            names (list[str]): List of class names.

        Returns:
            (nn.ModuleList): List of vocabulary modules.
        r~   )ru  r�   rl   rB   rÄ   r²   r   r  r#  r%  r´   rµ   r¶   rÌ   Úper8  r{  r%  r$  r   rV  rá  r“   )rp   r  r�  Útper¶   r$  rW  r\  s           rt   Ú	get_vocabzYOLOEModel.get_vocab7  s0  € ð ”=Ð Ð Ð ØŒz˜"Œ~ˆÝ˜$¥Ñ,Ô,Ð,Ð,Ð,Ø”=Ð Ð Ð ÝÕ& uÑ-Ô-×4Ò4Ñ6Ô6Ñ7Ô7ˆà×Ò˜uÑ%Ô%ˆØ×Ò˜ Ñ$Ô$Ð$Ý�d”j×+Ò+Ñ-Ô-Ñ.Ô.Ô5ˆØ�	Š	�$”'—*’*˜VÑ$Ô$Ñ%Ô%Ð%à"&¤,Ð<ˆdÔÐ°D´HˆÝ”‘”ˆØð 	'ð 	'ˆHÝ˜h­¬Ñ6Ô6Ð6Ð6Ð6Ø�LŠL˜ "œÑ&Ô&Ð&Ð&Øˆrv   c                óÞ   — t          | j        d         d¦  «        r
J d¦   «         ‚|j        dk    sJ ‚t          |¦  «        | _        || _        t          |¦  «        | j        d         _        dS )zÙSet classes in advance so that model could do offline-inference without clip model.

        Args:
            names (list[str]): List of class names.
            embeddings (torch.Tensor): Embeddings tensor.
        r~   rM  z^Prompt-free model does not support setting classes. Please try with Text/Visual prompt models.rg  N)rÆ   r�   r¶  r   r  r_  rî   rX  )rp   r  r�   s      rt   r%  zYOLOEModel.set_classesR  sv   € õ ˜4œ: bœ>¨6Ñ2Ô2ð 	
ð 	
Ølñ	
ô 	
Ð2ð Œ !Ò#Ð#Ð#Ð#Ý& uÑ-Ô-ˆŒ
ØˆŒÝ ™JœJˆŒ
�2ŒÔÐÐrv   c                ó6  — g }|�"|j         dk    sJ ‚|                     |¦  «         |�"|j         dk    sJ ‚|                     |¦  «         |s8|                     t          | dt          j        ddd¦  «        ¦  «        ¦  «         t          j        |d¬¦  «        S )a  Get class positional embeddings.

        Args:
            tpe (torch.Tensor | None): Text positional embeddings.
            vpe (torch.Tensor | None): Visual positional embeddings.

        Returns:
            (torch.Tensor): Class positional embeddings.
        Nrg  r_  r‰   r  r  rŠ   )r¶  r“   rÓ   r–   rv  r›   )rp   r`  rG  Úall_pes       rt   Ú
get_cls_pezYOLOEModel.get_cls_pea  sŸ   € ð ˆØˆ?Ø”8˜q’=�=�=�=Ø�MŠM˜#ÑÔÐØˆ?Ø”8˜q’=�=�=�=Ø�MŠM˜#ÑÔÐØð 	HØ�MŠM�' $¨­e¬k¸!¸RÀÑ.EÔ.EÑFÔFÑGÔGÐGÝŒy˜ QÐ'Ñ'Ô'Ð'rv   c                óx  ‡‡— g g g cŠ}}	‰j         d         }
|rt          |¦  «        ndh}t          |¦  «        }| j        D �]÷}|j        dk    r;t          |j        t          ¦  «        r‰|j                 nˆˆfd„|j        D ¦   «         Š|r|                      |‰|¦  «         t          |t          ¦  «        r³|�| 	                    ‰|¦  «        nd}|r|€J ‚| j
        rJ ‚|c S |                      |                     |¦  «        |¦  «                             ‰d         ¦  «        }|j         d         |
k    s|j        r|                     |
dd¦  «        }‰                     |¦  «          |‰¦  «        Š‰                     |j        | j        v r‰nd¦  «         |j        |v r”|	                     t&          j        j                             ‰d¦  «                             d¦  «                             d¦  «        ¦  «         |j        |k    r+t'          j        t'          j        |	d¦  «        d¬¦  «        c S �Œù‰S )a‹  Perform a forward pass through the model.

        Args:
            x (torch.Tensor): The input tensor.
            profile (bool): If True, profile the computation time for each layer.
            tpe (torch.Tensor, optional): Text positional embeddings.
            augment (bool): If True, perform data augmentation during inference.
            embed (list, optional): A list of layer indices to return embeddings from.
            vpe (torch.Tensor, optional): Visual positional embeddings.
            return_vpe (bool): If True, return visual positional embeddings.

        Returns:
            (torch.Tensor): Model's output tensor.
        r   r~   c                ó0   •— g | ]}|d k    r‰n‰|         ‘ŒS r€   r�   r‚   s     €€rt   r†   z&YOLOEModel.predict.<locals>.<listcomp>‹  r‡   rv   Nrˆ   r‰   rŠ   )rï   r�   rŽ   r�   r�   rl   r‘   r’   rB   Úget_vperu  re  rC  r5  r6  r7  r“   r”   r•   r–   r   r—   r˜   r™   rš   r›   )rp   rq   rz   r`  r{   r|   rG  rH  rœ   r�   Úbrž   rŸ   Úcls_per…   s    `            @rt   ro   zYOLOEModel.predictv  s>  øø€ ð   BÐˆˆ2ˆzØŒG�AŒJˆØ$)Ð3•	˜%Ñ Ô Ð °¨tˆÝ�e‘*”*ˆØ”ð 	Iñ 	IˆAØŒs�bŠyˆyÝ(¨¬­cÑ2Ô2Ð]�A�a”c”F�FÐ8]Ð8]Ð8]Ð8]Ð8]ÐYZÔY\Ð8]Ñ8]Ô8]�Øð 2Ø×'Ò'¨¨1¨bÑ1Ô1Ð1Ý˜!�[Ñ)Ô)ð 	!Ø+.¨?�a—i’i  3Ñ'Ô'Ð'À�Øð Ø˜?˜?˜?Ø#œ}Ð,Ð,Ð,Ø�J�J�JØŸš¨¯ª°3©¬¸Ñ=Ô=×EÒEÀaÈÄdÑKÔK�Ø”< ”? aÒ'Ð'¨1¬8Ð'Ø#Ÿ]š]¨1¨b°"Ñ5Ô5�FØ—’˜Ñ Ô Ð Ø��!‘”ˆAà�HŠH˜!œ# ¤Ð*Ð*�Q�Q°Ñ5Ô5Ð5ØŒs�eˆ|ˆ|Ø×!Ò!¥%¤(Ô"5×"IÒ"IÈ!ÈVÑ"TÔ"T×"\Ò"\Ð]_Ñ"`Ô"`×"hÒ"hÐikÑ"lÔ"lÑmÔmÐmØ”3˜'’>�>Ý œ<­¬	°*¸aÑ(@Ô(@ÀaÐHÑHÔHÐHÐHÐHùØˆrv   c                óÊ  — t          | d¦  «        spddlm} |                     dd¦  «        du}|r7t	          | j        d         dd¦  «        �t          | |¦  «        n || ¦  «        n|                      ¦   «         | _        |€L|  	                    |d         d|v rdn|                     d	d¦  «        |                     dd¦  «        ¬
¦  «        }|                      ||¦  «        S )rD  rE  r   )ÚTVPDetectLossÚvisualsNr~   rÑ   rF  r  )r`  rG  )
rÆ   Úultralytics.utils.lossrl  ré   rÓ   r�   rU   rG  rE  ru   )rp   rI  rJ  rl  Úvisual_prompts        rt   rn   zYOLOEModel.loss¡  s  € õ �t˜[Ñ)Ô)ð 	Ø<Ð<Ð<Ð<Ð<Ð<à!ŸIšI i°Ñ6Ô6¸dÐBˆMð !ð+å˜tœz¨"œ~¨}¸dÑCÔCÐOõ ˜D -Ñ0Ô0Ð0à&˜ tÑ,Ô,Ð,ð ×(Ò(Ñ*Ô*ð ŒNð ˆ=Ø—L’LØ�e”Ø%¨Ð.Ð.�D�D°E·I²I¸kÈ4Ñ4PÔ4PØ—I’I˜i¨Ñ.Ô.ð !ñ ô ˆEð
 �~Š~˜e UÑ+Ô+Ð+rv   )r=  rg  NT)r  FFr  )FNFNNF)r£   rO  rP  rQ  rs  rg   r#  rJ  r]  ra  r%  re  ro   rn   rR  rS  s   @rt   r<  r<  ´  s  ø€ € € € € ðð ð4
Hð 
Hð 
Hð 
Hð 
Hð 
Hð ÐÑÔð#'ð #'ð #'ñ Ôð#'ðJ ÐÑÔð
6ð 
6ñ Ôð
6ð'ð 'ð 'ð 'ðRð ð ð6'ð 'ð 'ð(ð (ð (ð*)ð )ð )ð )ðV,ð ,ð ,ð ,ð ,ð ,ð ,ð ,rv   r<  c                  ó0   ‡ — e Zd ZdZdˆ fd„	Zd	ˆ fd„	Zˆ xZS )
ÚYOLOESegModela+  YOLOE segmentation model.

    This class extends YOLOEModel to handle instance segmentation tasks with text and visual prompts, providing
    specialized loss computation for pixel-level object detection and segmentation.

    Methods:
        __init__: Initialize YOLOE segmentation model.
        loss: Compute loss with prompts for segmentation.

    Examples:
        Initialize a YOLOE segmentation model
        >>> model = YOLOESegModel("yoloe-v8s-seg.yaml", ch=3, nc=80)
        >>> results = model.predict(image_tensor, tpe=text_embeddings)
    úyoloe-v8s-seg.yamlrg  NTc                óR   •— t          ¦   «                              ||||¬¦  «         dS )aM  Initialize YOLOE segmentation model with given config and parameters.

        Args:
            cfg (str | dict): Model configuration file path or dictionary.
            ch (int): Number of input channels.
            nc (int, optional): Number of classes.
            verbose (bool): Whether to display model information.
        r¦  Nr§  r¨  s        €rt   rs  zYOLOESegModel.__init__Î  r©  rv   c                óH  •— t          | d¦  «        spddlm} |                     dd¦  «        du}|r7t	          | j        d         dd¦  «        �t          | |¦  «        n || ¦  «        n|                      ¦   «         | _        t          ¦   «          
                    ||¦  «        S )rD  rE  r   )ÚTVPSegmentLossrm  Nr~   rÑ   )rÆ   rn  ru  ré   rÓ   r�   rU   rG  rE  rÚ   rn   )rp   rI  rJ  ru  ro  r¢   s        €rt   rn   zYOLOESegModel.lossÙ  s·   ø€ õ �t˜[Ñ)Ô)ð 	Ø=Ð=Ð=Ð=Ð=Ð=à!ŸIšI i°Ñ6Ô6¸dÐBˆMð !ð+å˜tœz¨"œ~¨}¸dÑCÔCÐOõ ˜D .Ñ1Ô1Ð1à'˜¨Ñ-Ô-Ð-ð ×(Ò(Ñ*Ô*ð ŒNõ ‰wŒw�|Š|˜E 5Ñ)Ô)Ð)rv   )rr  rg  NTr  )r£   rO  rP  rQ  rs  rn   rR  rS  s   @rt   rq  rq  ¾  sk   ø€ € € € € ðð ð	Að 	Að 	Að 	Að 	Að 	Að*ð *ð *ð *ð *ð *ð *ð *ð *ð *rv   rq  c                  ó*   ‡ — e Zd ZdZˆ fd„Zdd„Zˆ xZS )ÚEnsemblea  Ensemble of models.

    This class allows combining multiple YOLO models into an ensemble for improved performance through model averaging
    or other ensemble techniques.

    Methods:
        __init__: Initialize an ensemble of models.
        forward: Generate predictions from all models in the ensemble.

    Examples:
        Create an ensemble of models
        >>> ensemble = Ensemble()
        >>> ensemble.append(model1)
        >>> ensemble.append(model2)
        >>> results = ensemble(image_tensor)
    c                óH   •— t          ¦   «                              ¦   «          dS )z!Initialize an ensemble of models.Nr§  )rp   r¢   s    €rt   rs  zEnsemble.__init__  s   ø€ å‰Œ×ÒÑÔÐÐÐrv   Fc                óZ   ‡‡‡— ˆˆˆfd„| D ¦   «         }t          j        |d¦  «        }|dfS )aŽ  Run ensemble forward pass and concatenate predictions from all models.

        Args:
            x (torch.Tensor): Input tensor.
            augment (bool): Whether to augment the input.
            profile (bool): Whether to profile the model.

        Returns:
            (torch.Tensor): Concatenated predictions from all models.
            (None): Always None for ensemble inference.
        c                ó6   •— g | ]} |‰‰‰¦  «        d          ‘ŒS )r   r�   )rƒ   Úmoduler{   rz   rq   s     €€€rt   r†   z$Ensemble.forward.<locals>.<listcomp>  s,   ø€ Ð?Ð?Ð?°ˆVˆV�A�w Ñ(Ô(¨Ô+Ð?Ð?Ð?rv   r§   N)r–   r›   )rp   rq   r{   rz   r…   s    ``` rt   ru   zEnsemble.forward  sA   øøø€ ð @Ð?Ð?Ð?Ð?Ð?¸$Ð?Ñ?Ô?ˆõ ŒI�a˜‰OŒOˆØ�$ˆwˆrv   )FF)r£   rO  rP  rQ  rs  ru   rR  rS  s   @rt   rw  rw  ñ  sV   ø€ € € € € ðð ð"ð ð ð ð ðð ð ð ð ð ð ð rv   rw  c              #  ót  K  — | €i } |€i }ddl }ddlm} t          ¦   «         }g }t          5  	 |                     ¦   «         D ]›\  }}|                     dd¦  «        \  }}	|                     dd¦  «        \  }
} ||¦  «        }|                     ||	|j         	                    |	|¦  «        f¦  «         t          ||	t           ||
¦  «        |¦  «        ¦  «         Œœ|                      ¦   «         D ]\  }} ||¦  «        |j        |<   ŒdV — | D ]}||j        v r|j        |= Œ|D ],\  }}}||u rt          ||¦  «         Œt          |||¦  «         Œ-nJ# | D ]}||j        v r|j        |= Œ|D ],\  }}}||u rt          ||¦  «         Œt          |||¦  «         Œ-w xY w	 ddd¦  «         dS # 1 swxY w Y   dS )a‹  Context manager for temporarily adding or modifying modules in Python's module cache (`sys.modules`).

    This function can be used to change the module paths during runtime. It's useful when refactoring code, where you've
    moved a module from one location to another, but you still want to support the old import paths for backwards
    compatibility.

    Args:
        modules (dict, optional): A dictionary mapping old module paths to new module paths.
        attributes (dict, optional): A dictionary mapping old module attributes to new module attributes.

    Examples:
        >>> with temporary_modules({"old.module": "new.module"}, {"old.module.attribute": "new.module.attribute"}):
        >>> import old.module  # this will now import new.module
        >>> from old.module import attribute  # this will now import new.module.attribute

    Notes:
        The changes are only in effect inside the context manager and are undone once the context manager exits.
        Be aware that directly manipulating `sys.modules` can lead to unpredictable results, especially in larger
        applications or libraries. Use this function with caution.
    Nr   )Úimport_moduler)  r‰   )ÚsysÚ	importlibr}  ÚobjectÚ_temporary_modules_lockr  Úrsplitr“   Ú__dict__ré   r€  rÓ   rÅ   rÉ   )rÅ   Ú
attributesr~  r}  ÚmissingÚpreviousÚoldÚnewÚ
old_moduleÚold_attrÚ
new_moduleÚnew_attrr{  r,  r}  s                  rt   Útemporary_modulesr�     s   è è € ð, €ØˆØÐØˆ
Ø€J€J€JØ'Ð'Ð'Ð'Ð'Ð'å‰hŒh€GØ€HÝ	 ð 1ð 1ð	1à&×,Ò,Ñ.Ô.ð Xð X‘��SØ'*§z¢z°#°qÑ'9Ô'9Ñ$�
˜HØ'*§z¢z°#°qÑ'9Ô'9Ñ$�
˜HØ&˜ zÑ2Ô2�Ø—’ ¨°6´?×3FÒ3FÀxÐQXÑ3YÔ3YÐ ZÑ[Ô[Ð[Ý˜ ­'°-°-À
Ñ2KÔ2KÈXÑ*VÔ*VÑWÔWÐWÐWð $ŸMšM™OœOð 6ð 6‘��SØ#0 =°Ñ#5Ô#5�”˜CÑ Ð àˆEˆEˆEð ð )ð )�Ø˜#œ+Ð%Ð%Øœ CÐ(øØ'/ð 1ð 1Ñ#�˜˜eØ˜GÐ#Ð#Ý˜F DÑ)Ô)Ð)Ð)å˜F D¨%Ñ0Ô0Ð0Ð0ð	1øð ð )ð )�Ø˜#œ+Ð%Ð%Øœ CÐ(øØ'/ð 1ð 1Ñ#�˜˜eØ˜GÐ#Ð#Ý˜F DÑ)Ô)Ð)Ð)å˜F D¨%Ñ0Ô0Ð0Ð0ð	1øøøð 1ð)1ð 1ð 1ñ 1ô 1ð 1ð 1ð 1ð 1ð 1ð 1ð 1øøøð 1ð 1ð 1ð 1ð 1ð 1s+   ¬F-®C!EÄAF-ÅAFÆF-Æ-F1Æ4F1c                  óæ   — e Zd ZdZ eej        d¦  «        ZdZ e	j
        ¦   «         Z e	j        ¦   «         Zed„ ¦   «         Zeej        d„ ¦   «         ¦   «         Zed„ ¦   «         Zed„ ¦   «         ZdS )Ú	_SafeLoadaè  Opt-in restricted checkpoint loading: reconstruct only known model classes (`weights_only=True` plus an
    allow-list) and build models without `eval()`.

    Enabled per-process by the `ULTRALYTICS_SAFE_LOAD` env flag, or per-call by `torch_safe_load(..., safe_only=True)`.
    Default loading (flag off) is unchanged. The globals a restricted load registers stay registered for the process, so
    they also apply to any other `torch.load(weights_only=True)` call made afterwards.
    Ú get_unsafe_globals_in_checkpointNc                óJ   — | j         ot          pt          | j        dd¦  «        S )zfWhether model construction should use the no-eval, known-layer path (env flag or an in-progress load).ÚactiveF)Ú	SUPPORTEDrH   rÓ   Ú_local)r”  s    rt   Ú
restrictedz_SafeLoad.restrictedl  s$   € ð Œ}ÐT¥)Ð"S­w°s´zÀ8ÈUÑ/SÔ/SÐTrv   c           	   #  ó\  ‡ K  — 	 t           j                             |¦  «        }n# t          $ r g }Y nw xY w‰ j        5  ‰ j        €‰                      ¦   «         ‰ _        |D ]µ}|                     d¦  «        \  }}}|‰ j        vr‘|dv s!|                     d¦  «        d         dv s|dv rlt          t          j
        |¦  «        |d¦  «        }t          |t          ¦  «        r4|dv r|j        |k    r%nt          |t          j        ¦  «        r
|‰ j        |<   Œ¶t#          d„ |D ¦   «         ¦  «        rLddlmc m} ddlm}	 |j        |j        |j        |j        |j        |	fD ]}|‰ j        |j        › d|j        › �<   Œd	|v rvddl}
dd
lm} ||
j        j        |
j        j         |
j        j!        |
j        j"        |
j        j#        |
j        j$        |
j        j%        fD ]}|‰ j        |j        › d|j        › �<   Œˆ fd„|D ¦   «         }d|v r>| &                    t          tO          j(        tN          j)        ¦  «        ¦  «        ¦  «         |rt           j         *                    |¦  «         ddd¦  «         n# 1 swxY w Y   d‰ j+        _,        	 dV — d‰ j+        _,        dS # d‰ j+        _,        w xY w)ug  Load with `weights_only=True`: register the globals this checkpoint needs and mark the thread restricted, so
        a checkpoint that reaches model construction (parse_model) also uses the no-eval, known-layer path.

        Globals are registered with `add_safe_globals` for the life of the process, never scoped per load: the
        `safe_globals()` context manager removes its entries from a process-global set on exit, so with concurrent
        loads one thread's exit strips the allow-list out of another thread's in-flight unpickle. Registering only
        the globals a checkpoint references also keeps the restricted unpickler fast â€” torch rebuilds its lookup from
        the whole registered set on every GLOBAL/NEWOBJ/REDUCE/BUILD opcode, so a 660-entry allow-list nearly doubled
        the load time of a checkpoint that references 20 of them.
        Nr)  >   úultralytics.nn.tasksútorch.nn.modulesúultralytics.nn.modulesr   >   r˜  r™  >   úultralytics.utils.talúultralytics.utils.lossc              3  ó@   K  — | ]}|                      d ¦  «        V — ŒdS )ztorchvision.transforms.N)Ú
startswith)rƒ   r+  s     rt   rÔ   z$_SafeLoad.loading.<locals>.<genexpr>“  s/   è è € ÐQÐQÀ$�4—?’?Ð#<Ñ=Ô=ÐQÐQÐQÐQÐQÐQrv   )ÚInterpolationModezultralytics.nn.text_model.CLIP)ÚCLIPc                ó@   •— g | ]}|‰j         v ¯‰j         |         |f‘ŒS r�   )Ú	_registry)rƒ   r+  r”  s     €rt   r†   z%_SafeLoad.loading.<locals>.<listcomp>ª  s3   ø€ Ð_Ð_Ð_°tÈÐQTÔQ^ÐI^ÐI^˜œ dÔ+¨TÐ2ÐI^ÐI^ÐI^rv   znumpy.dtypeTF)-r–   Úserializationr�  rÙ  Ú_lockr¡  Ú_buildÚ
rpartitionrÓ   r  r}  rl   rº   rO  Ú
issubclassr   ÚModulerÕ   Ú!torchvision.transforms.transformsÚ
transformsÚ!torchvision.transforms.functionalrž  ÚComposeÚ	NormalizeÚResizeÚ
CenterCropÚToTensorrP  Úclipr/  rŸ  r�   Ú	LayerNormÚ	QuickGELUÚResidualAttentionBlockÚTransformerÚVisionTransformerÚ_convert_image_to_rgbr“   r¹   r"  Úfloat64Úadd_safe_globalsr”  r’  )r”  r'  Úneededr+  r{  r¿   r,  ÚobjÚtvtrž  r°  rŸ  Úentriess   `            rt   Úloadingz_SafeLoad.loadingq  su  øè è € ð	ÝÔ(×IÒIÈ&ÑQÔQˆFˆFøÝð 	ð 	ð 	ØˆFˆFˆFð	øøøàŒYð ,	>ð ,	>ØŒ}Ð$Ø #§
¢
¡¤�”Øð 2ð 2�Ø"&§/¢/°#Ñ"6Ô"6‘�˜˜4Ø˜sœ}Ð,Ð,ØÐdÐdÐdØ×(Ò(¨Ñ-Ô-¨aÔ0Ð4bÐbÐbØÐ!TÐTÐTå!¥)Ô"9¸&Ñ"AÔ"AÀ4ÈÑNÔN�CÝ! #¥tÑ,Ô,ð 2à!Ð%XÐXÐXð œ¨&Ò0Ð0Ð0å'¨­R¬YÑ7Ô7ð 1ð /2˜œ dÑ+øÝÐQÐQÈ&ÐQÑQÔQÑQÔQð Pà?Ð?Ð?Ð?Ð?Ð?Ð?Ð?Ð?ØOÐOÐOÐOÐOÐOàœK¨¬¸¼
ÀCÄNÐTWÔT`ÐbsÐtð Pð P�CØLO�C”M S¤^Ð"HÐ"H°cÔ6FÐ"HÐ"HÑIÐIØ/°6Ð9Ð9Ø���à:Ð:Ð:Ð:Ð:Ð:ð Ø”J”OØ”JÔ(Ø”JÔ(Ø”JÔ5Ø”JÔ*Ø”JÔ0Ø”IÔ3ð	ð 
Pð 
P�Cð MP�C”M S¤^Ð"HÐ"H°cÔ6FÐ"HÐ"HÑIÐIØ_Ð_Ð_Ð_¸vÐ_Ñ_Ô_ˆGØ Ð&Ð&Ø—’�t¥B¤H­R¬ZÑ$8Ô$8Ñ9Ô9Ñ:Ô:Ð:Øð >ÝÔ#×4Ò4°WÑ=Ô=Ð=ðY,	>ð ,	>ð ,	>ñ ,	>ô ,	>ð ,	>ð ,	>ð ,	>ð ,	>ð ,	>ð ,	>øøøð ,	>ð ,	>ð ,	>ð ,	>ðZ !ˆŒ
Ôð	&ØˆEˆEˆEà %ˆCŒJÔÐÐø ˆCŒJÔÐ%Ð%Ð%Ð%s*   …% ¥4³4¾H)I3É3I7É:I7ÊJ ÊJ+c                ó2  ‡— ddl Š	 ‰                     |                      ¦   «         d¬¦  «        j        }t	          |‰j        ¦  «        sJ ‚g }|j        }t	          |‰j        ¦  «        r6|                     |j	        ¦  «         |j
        }t	          |‰j        ¦  «        °6t	          |‰j        ¦  «        sJ ‚|                     |j        ¦  «         |dd…         dgddgfv s
J d¦   «         ‚t          t          |d         ¦  «        }t	          |t          ¦  «        rt!          |t          j        ¦  «        sJ ‚ˆfd	„|j        D ¦   «         }ˆfd
„|j        D ¦   «         } ||i |¤ŽS # t(          $ r&}t+          t-          d| › d�¦  «        ¦  «        |‚d}~ww xY w)a  Resolve a model-yaml `activation` spec to a `torch.nn` module instance without `eval()`.

        Accepts only the documented `[torch.]nn.<Class>(literal args)` shape (e.g. `nn.SiLU()`,
        `torch.nn.LeakyReLU(0.1)`) with literal arguments, and rejects anything else.
        r   Nrt  )Úmoder‰   r   r–   z#activation must be a torch.nn classc                ó:   •— g | ]}‰                      |¦  «        ‘ŒS r�   )Úliteral_eval)rƒ   ÚaÚasts     €rt   r†   z(_SafeLoad.activation.<locals>.<listcomp>Ë  s'   ø€ Ð;Ð;Ð;¨A�C×$Ò$ QÑ'Ô'Ð;Ð;Ð;rv   c                óP   •— i | ]"}|j         ‰                     |j        ¦  «        “Œ#S r�   )ÚargrÁ  r}  )rƒ   ÚkwrÃ  s     €rt   r  z(_SafeLoad.activation.<locals>.<dictcomp>Ì  s-   ø€ ÐQÐQÐQ¸R�b”f˜c×.Ò.¨r¬xÑ8Ô8ÐQÐQÐQrv   u%   ERROR â�Œï¸� unsupported activation 'z'' blocked during restricted model load.)rÃ  Úparser
  Úbodyrl   ÚCallÚfuncÚ	Attributer“   r,  r}  ÚNameÚidrÓ   r   rº   r¦  r§  rr   ÚkeywordsÚ	ExceptionÚ	TypeErrorrN   )	ÚactÚcallÚattrsÚnodeÚklassrr   rs   r�  rÃ  s	           @rt   Ú
activationz_SafeLoad.activationµ  sÇ  ø€ ð 	ˆ
ˆ
ˆ
ð	Ø—9’9˜SŸYšY™[œ[¨v�9Ñ6Ô6Ô;ˆDÝ˜d C¤HÑ-Ô-Ð-Ð-Ð-ØˆEØ”9ˆDÝ˜T 3¤=Ñ1Ô1ð "Ø—’˜TœYÑ'Ô'Ð'Ø”z�õ ˜T 3¤=Ñ1Ô1ð "õ ˜d C¤HÑ-Ô-Ð-Ð-Ð-Ø�LŠL˜œÑ!Ô!Ð!Ø˜˜˜”9 $ ¨$°¨Ð 9Ð9Ð9Ð9Ð;`Ñ9Ô9Ð9Ý�B  a¤Ñ)Ô)ˆEÝ˜e¥TÑ*Ô*ÐK­z¸%ÅÄÑ/KÔ/KÐKÐKÐKØ;Ð;Ð;Ð;°´Ð;Ñ;Ô;ˆDØQÐQÐQÐQÀ4Ä=ÐQÑQÔQˆFØ�5˜$Ð) &Ð)Ð)Ð)øÝð 	ð 	ð 	ÝÝÐk¸sÐkÐkÐkÑlÔlñô àðøøøøð	øøøs   ‡EE& Å&
FÅ0!FÆFc                ó†  ‡— ddl Šddl}ddlmc m} g }t          j        d¦  «                             ¦   «         d         }|t
          j        |df|dfgz  }| 	                    t          ¦  «         | 	                    t          df¦  «         ˆfd„}|t          j        dft          df|j        d	f|d
fgz  }t          r=||j        |j        df|j        df|j        |j        j        › d|j        j        › �fgz  }n<||j        |j        df|j        df|j        |j        j        › d|j        j        › �fgz  }d„ |D ¦   «         S )zaBuild the known data globals and legacy aliases; model classes are resolved only when referenced.r   Nznumpy.core.multiarray.scalarznumpy._core.multiarray.scalarz.ultralytics.yolo.utils.IterableSimpleNamespacec                ó°  •— t          | t          ¦  «        rF|                     d¦  «        s1t          | t          j        ‰j        f¦  «        rt          | |¦  «        S t          | t          j        ¦  «        r4|dv r0t          t          | ¦  «        |¦  «                             | ¦  «        S t          j
        dt          | ¦  «        j        › d|›d�¦  «        ‚)NÚ__>   ru   rÊ   zunsafe getattr(z, z&) blocked during restricted model load)rl   rº   r�  r¦  r   r§  ÚEnumrÓ   Ú__get__ÚpickleÚUnpicklingErrorr£   )rº  r+  Úenums     €rt   Ú_getattrz"_SafeLoad._build.<locals>._getatträ  sÎ   ø€ Ý˜#�tÑ$Ô$ð *¨T¯_ª_¸TÑ-BÔ-Bð *ÅzÐRUÕXZÔXaÐcgÔclÐWmÑGnÔGnð *Ý˜s DÑ)Ô)Ð)Ý˜#�rœyÑ)Ô)ð =¨dÐ6QÐ.QÐ.QÝ�t C™yœy¨$Ñ/Ô/×7Ò7¸Ñ<Ô<Ð<ÝÔ(Øf¥$ s¡)¤)Ô"4ÐfÐf¸ÐfÐfÐfñô ð rv   ú$ultralytics.nn.modules.block.Silenceú*ultralytics.nn.tasks.YOLOv10DetectionModelú$ultralytics.utils.loss.v10DetectLosszbuiltins.getattrúpathlib.WindowsPathúpathlib.PosixPathr)  c                ó¦   — i | ]N}t          |t          ¦  «        r|d          n|j        › d|j        › �t          |t          ¦  «        r|d         n|“ŒOS )r‰   r)  r   )rl   ÚtuplerO  rP  )rƒ   r�  s     rt   r  z$_SafeLoad._build.<locals>.<dictcomp>  sm   € ð 
ð 
ð 
ð õ   ¥5Ñ)Ô)ÐQˆQˆqŒTˆT°!´,Ð/QÐ/QÀÄÐ/QÐ/QÝ" 1¥eÑ,Ô,Ð3��!”�°!ð
ð 
ð 
rv   )rÞ  Úpathlibrn  Úutilsrn   r¹   r·  Ú
__reduce__r"  r“   rL   r   ÚIdentityre  ÚE2EDetectLossrJ   ÚWindowsPathÚ	PosixPathrO  rP  )r”  rç  Úul_lossÚallowÚscalarrß  rÞ  s         @rt   r¤  z_SafeLoad._buildÓ  sÊ  ø€ ð 	ˆˆˆØˆˆˆà0Ð0Ð0Ð0Ð0Ð0Ð0Ð0Ð0àˆõ ”˜A‘”×)Ò)Ñ+Ô+¨AÔ.ˆØ•"”(˜VÐ%CÐDÀvÐOnÐFoÐpÑpˆØ�ŠÕ,Ñ-Ô-Ð-Ø�ŠÕ-Ð/_Ð`ÑaÔaÐað	ð 	ð 	ð 	ð 	ð 	ÝŒ[Ð@ÐAÝÐIÐJØÔ"Ð$JÐKØÐ)Ð*ð	
ñ 	
ˆõ ð 	ØØÔ#ØÔ$Ð&;Ð<ØÔ$Ð&9Ð:ØÔ$¨Ô):Ô)EÐ&hÐ&hÈÔHYÔHfÐ&hÐ&hÐið	ñ ˆEˆEð ØÔ!ØÔ"Ð$7Ð8ØÔ"Ð$9Ð:ØÔ" wÔ':Ô'EÐ$jÐ$jÈÔH[ÔHhÐ$jÐ$jÐkð	ñ ˆEð
ð 
ð ð	
ñ 
ô 
ð 	
rv   )r£   rO  rP  rQ  rÆ   r–   r¢  r“  r¡  Ú	threadingÚLockr£  Úlocalr”  Úclassmethodr•  Ú
contextlibÚcontextmanagerr½  r¡  rÖ  r¤  r�   rv   rt   r�  r�  Z  sÜ   € € € € € ðð ð �˜Ô+Ð-OÑPÔP€IØ€Iàˆ	ŒÑÔð 
ð ˆYŒ_ÑÔ€FàðUð Uñ „[ðUð ØÔð@&ð @&ñ Ôñ „[ð@&ðD ðð ñ „\ðð: ð2
ð 2
ñ „[ð2
ð 2
ð 2
rv   r�  c                óÜ  ‡‡	— ddl m}m} ‰€t          Š‰rt          j        sdŠt          | d¬¦  «          || ¦  «        Š	ˆ	ˆfd„}t          d| › d	�¦  «        }	  |¦   «         }�n # t          t          t          j        f$ �ri}t          |t          ¦  «        r!d
t          |¦  «        v rt          |¦  «        |‚t          |t          ¦  «        rdt          |¦  «        vr‚ ‰r;t          |t          j        ¦  «        r!t          t          d| › d�¦  «        ¦  «        |‚t          t          | ¦  «        ¦  «        j        }t          | ¦  «        |k    s||vr9t          t          d| › dt#          |¦  «        j        › d|› d�¦  «        ¦  «        |‚t'          j        d‰	› d| › d�¦  «         t          ‰	¦  «                             d¬¦  «          || ¦  «        Š	 |¦   «         }Y d}~�nd}~wt,          $ �r}|j        dv r!t          t          d| › d�¦  «        ¦  «        |‚|j        dk    r6t-          t          d| › dt/          d¦  «        j        › d�¦  «        ¦  «        |‚|j        rC|j                             d¦  «        r)t-          t          d| › d|j        › d�¦  «        ¦  «        |‚‰r‚ t'          j        | › d|j        › d|j        › d �¦  «         t5          |j        ¦  «         t7          ‰	d!¬"¦  «        }Y d}~nd}~ww xY wt          |t8          j        j        ¦  «        rt          |¦  «        ‚t          |t>          ¦  «        s!t'          j        d#| › d$�¦  «         d%|j         i}|‰	fS )&a¼  Attempt to load a PyTorch model with the torch.load() function. If a ModuleNotFoundError is raised, it catches
    the error, logs a warning message, and attempts to install the missing module via the check_requirements()
    function. After installation, the function again attempts to load the model using torch.load().

    Args:
        weight (str | Path): The file path of the PyTorch model.
        safe_only (bool, optional): Load with `torch.load(weights_only=True)`, reconstructing only the known
            Ultralytics/torch model classes on the allow-list. Defaults to the `ULTRALYTICS_SAFE_LOAD` environment
            variable (off), so standard usage is unchanged; set the env to opt in.

    Returns:
        (dict): The loaded model checkpoint.
        (str): The loaded filename.

    Examples:
        >>> from ultralytics.nn.tasks import torch_safe_load
        >>> ckpt, file = torch_safe_load("path/to/best.pt", safe_only=True)
    r   )ÚGITHUB_ASSETS_NAMESÚattempt_download_assetNFz.pt)ÚfileÚsuffixc                 óN  •— t          ddddœddddœt          rd	d
ind
d	i¥¬¦  «        5  ‰rPt                               ‰ ¦  «        5  t	          ‰ dd¬¦  «        cd d d ¦  «         cd d d ¦  «         S # 1 swxY w Y   t	          ‰ d¬¦  «        cd d d ¦  «         S # 1 swxY w Y   d S )Nzultralytics.utilszultralytics.models.yolozultralytics.data)zultralytics.yolo.utilszultralytics.yolo.v8zultralytics.yolo.dataztorch.nn.Identityz#ultralytics.nn.tasks.DetectionModelz$ultralytics.utils.loss.E2EDetectLoss)rà  rá  râ  rä  rã  )rÅ   r„  ÚcpuT)Úmap_locationÚweights_only©rþ  )r�  rJ   r�  r½  r^   )rú  Ú	safe_onlys   €€rt   Ú_loadztorch_safe_load.<locals>._load%  s™  ø€ Ýà*=Ø'@Ø);ðð ð 9LØ>cØ8^ð
ð 
õ ðFÐ(Ð*?Ð@Ð@à/Ð1DÐEð
ð
ñ 
ô 
ð 	8ð 	8ð$ ð SÝ×&Ò& tÑ,Ô,ð Sð SÝ% d¸ÈTÐRÑRÔRðSð Sð Sð Sñ Sô Sð Sð'	8ð 	8ð 	8ð 	8ñ 	8ô 	8ð 	8ð 	8ð&Sð Sð Sð Søøøð Sð Sð Sð Så˜d°Ð7Ñ7Ô7ð+	8ð 	8ð 	8ð 	8ñ 	8ô 	8ð 	8ð 	8ð 	8ð 	8ð 	8ð 	8øøøð 	8ð 	8ð 	8ð 	8ð 	8ð 	8s5   ªBÁA2ÁBÁ2A6	Á6BÁ9A6	Á:BÂBÂ!Bõ   ERROR â�Œï¸� z§ is a TorchScript archive, not an Ultralytics PyTorch checkpoint.
Load the original .pt weights, or export again with format='torchscript' and load that file directly.zTorchScript archiveÚPytorchStreamReaderúŽ references types outside the supported Ultralytics checkpoint format. Use an official Ultralytics model, i.e. 'yolo predict model=yolo26n.pt'uM    is not a loadable checkpoint â€” the file is empty, truncated or corrupted (z: z™).
Recommend fixes are to re-download or re-export the file, or to run a command with an official Ultralytics model, i.e. 'yolo predict model=yolo26n.pt'zCorrupt cache z, re-downloading z...T)Ú
missing_ok>   úmodels.yoloúmodels.commonúmodels.experimentalÚmodelsaw   appears to be an Ultralytics YOLOv5 model originally trained with https://github.com/ultralytics/yolov5. This model is NOT forwards compatible with YOLOv8 at https://github.com/ultralytics/ultralytics.
Recommend fixes are to train a new model using the latest 'ultralytics' package or to run a command with an official Ultralytics model, i.e. 'yolo predict model=yolo26n.pt'znumpy._corez( requires numpy>=1.26.1, however numpy==Únumpyz is installed.zultralytics.z& requires missing Ultralytics module 'z•'. Train a new model using the latest 'ultralytics' package or run a command with an official Ultralytics model, i.e. 'yolo predict model=yolo26n.pt'z appears to require 'zK', which is not in Ultralytics requirements.
AutoInstall will run now for 'zÞ' but this feature will be removed in the future.
Recommend fixes are to train a new model using the latest 'ultralytics' package or to run a command with an official Ultralytics model, i.e. 'yolo predict model=yolo26n.pt'rý  r   z
The file 'z€' appears to be improperly saved or formatted. For optimal results, use model.save('filename.pt') to correctly save YOLO models.r�   )!Úultralytics.utils.downloadsrø  rù  rH   r�  r“  rR   rN   ÚRuntimeErrorÚEOFErrorrÜ  rÝ  rl   r	  rÐ  r   r+  rº   r£   rG   r¡   ÚunlinkÚModuleNotFoundErrorÚ
__import__Ú__version__r�  rQ   r^   r–   ÚjitÚScriptModulerm   r�   )
r'  r  rø  rù  r  Útorchscript_errorÚckptr�  r+  rú  s
    `       @rt   Útorch_safe_loadr  	  sç  øø€ ð& XÐWÐWÐWÐWÐWÐWÐWàÐÝˆ	Øð �Ô,ð Øˆ	Ý�f UÐ+Ñ+Ô+Ð+Ø!Ð! &Ñ)Ô)€Dð8ð 8ð 8ð 8ð 8ð 8õ2 ð	q˜ð 	qð 	qð 	qñô Ðð
G4Øˆu‰wŒwˆ‰øå�(¥FÔ$:Ð;ð ñ ð õ
 �a�Ñ&Ô&ð 	6Ð+@ÅCÈÁFÄFÐ+JÐ+JÝÐ-Ñ.Ô.°AÐ5Ý�a�Ñ&Ô&ð 	Ð+@ÍÈAÉÌÐ+NÐ+NØØð 	� A¥vÔ'=Ñ>Ô>ð 	åÝð_ Fð _ð _ð _ñô ñô ð
 ðõ •C˜‘K”KÑ Ô Ô%ˆÝˆv‰;Œ;˜$ÒÐ $Ð.AÐ"AÐ"AÝÝðn Fð nð nÝ˜Q™œÔ(ðnð nØ,-ðnð nð nñô ñô ð ðõ 	ŒÐJ¨ÐJÐJ¸vÐJÐJÐJÑKÔKÐKÝˆT‰
Œ
×Ò TÐÑ*Ô*Ð*Ø%Ð% fÑ-Ô-ˆØˆu‰wŒwˆˆˆˆˆ‰øøøøåð $4ñ $4ð $4ØŒ6ÐVÐVÐVÝÝðn Fð nð nð nñô ñô ð ðð ŒV�}Ò$Ð$Ý%Ýð D Fð  Dð  DÕT^Ð_fÑTgÔTgÔTsð  Dð  Dð  Dñô ñô ð ð	ð
 ŒVð 	˜œ×)Ò)¨.Ñ9Ô9ð 	Ý%ÝðN Fð Nð NÐRSÔRXð Nð Nð Nñô ñô ð ðð ð 	ð ÝŒØð fð f¨A¬Fð fð fØ/0¬vðfð fð fñ	
ô 	
ð 	
õ 	˜1œ6Ñ"Ô"Ð"Ý˜$¨UÐ3Ñ3Ô3ˆˆˆˆˆˆøøøøðI$4øøøõL �$�œ	Ô.Ñ/Ô/ð +ÝÐ)Ñ*Ô*Ð*å�d�DÑ!Ô!ð %åŒða˜ð að að añ	
ô 	
ð 	
ð ˜œÐ$ˆà�ˆ:Ðs&   Á
A& Á&LÂEG'Ç'LÇ5DK>Ë>LTFc                óš  — t          | ¦  «                             ¦   «                              t          ¦  «        rt	          | t
          d         ¬¦  «        } t          | ¦  «        \  }} i t          ¥|                     di ¦  «        ¥}|                     d¦  «        p|                     d¦  «        }t          |t          j        j        ¦  «        s t          t          d| › d�¦  «        ¦  «        ‚|                     ¦   «         }|                     d¦  «        rt!          ||d         ¦  «         ||_        t          | ¦  «        |_        t'          |d	t)          |¦  «        ¦  «        |_        t-          |d
¦  «        st          j        dg¦  «        |_        |r$t-          |d¦  «        r|                     ¦   «         n|                     ¦   «                              |¦  «        }|                     ¦   «         D ]P}t-          |d¦  «        r||_        Œt          |t          j        j        ¦  «        rt-          |d¦  «        sd|_        ŒQ||fS )af  Load single model weights.

    Args:
        weight (str | Path): Model weight path.
        device (torch.device, optional): Device to load model to.
        inplace (bool): Whether to do inplace operations.
        fuse (bool): Whether to fuse model.

    Returns:
        (torch.nn.Module): Loaded model.
        (dict): Model checkpoint dictionary.
    Úweights_dir)Údownload_dirÚ
train_argsrã   r�   r  r  ÚmodeloptÚtaskrÜ   g      @@rÌ   r^  Úrecompute_scale_factorN) r	  r  r�  rO   rP   rI   r  rF   ré   rl   r–   r   r§  rÐ  rN   rê   re   rr   rå   rÓ   Úguess_model_taskr  rÆ   r2  rÜ   rÌ   rt  r8  rÅ   r^  ÚUpsampler  )	r'  r¶   r^  rÌ   r  rr   Ú	candidater�   rŸ   s	            rt   Úload_checkpointr"  š  s&  € õ ˆ6�{„{×ÒÑÔ×%Ò%Õ&:Ñ;Ô;ð JÝ˜Fµ¸-Ô1HÐIÑIÔIˆÝ" 6Ñ*Ô*�L€Dˆ&Ø?ÕÐ? 4§8¢8¨L¸"Ñ#=Ô#=Ð?€DØ—’˜‘”Ð4 4§8¢8¨GÑ#4Ô#4€IÝ�i¥¤¤Ñ1Ô1ð 
ÝÝð[ ð [ð [ð [ñô ñ
ô 
ð 	
ð �OŠOÑÔ€EØ‡x‚x�
ÑÔð -Ý�E˜4 
Ô+Ñ,Ô,Ð,ð €E„JÝ˜‘K”K€E„MÝ˜ Õ(8¸Ñ(?Ô(?Ñ@Ô@€E„JÝ�5˜(Ñ#Ô#ð ,Ý”| T FÑ+Ô+ˆŒà!ÐG¥g¨e°VÑ&<Ô&<ÐGˆU�ZŠZ‰\Œ\ˆ\À%×MÒMÑOÔO×RÒRÐSYÑZÔZ€Eð �]Š]‰_Œ_ð ,ð ,ˆÝ�1�iÑ Ô ð 	,ØˆAŒIˆIÝ˜�5œ8Ô,Ñ-Ô-ð 	,µg¸aÐAYÑ6ZÔ6Zð 	,Ø'+ˆAÔ$øð �$ˆ;Ðrv   c                ó"  ‡ ‡‡ ‡!‡"— ddl }d}t          d¦  «        }ˆ fd„dD ¦   «         \  }}}}	‰                      dd¦  «        }
ˆ fd	„d
D ¦   «         \  }}}‰                      d¦  «        }|rT|sFt          t	          |                     ¦   «         ¦  «        ¦  «        }t          j        d|› d�¦  «         ||         \  }}}t           	                    ¦   «         }|r]|rt           
                    |¦  «        nt          |¦  «        t          _        |r&t          j        t          d¦  «        › d|› �¦  «         |r(t          j        ddd›dd›dd›dd›ddd›dd›�¦  «         ‰gŠg g ‰d         }}}t!          h t"          ’t          ’t$          ’t&          ’t(          ’t*          ’t,          ’t.          ’t0          ’t2          ’t4          ’t6          ’t8          ’t:          ’t<          ’t>          ’t@          ’tB          ’tD          ’tF          ’tH          ’tJ          ’tL          ’tN          ’tP          ’tR          ’tT          j+        j,        ’tZ          ’t\          ’t^          ’t`          ’tb          ’td          ’tf          ’¦  «        }t!          t8          t:          t<          t>          t@          tL          tN          tP          tR          t\          t^          t0          td          t2          tf          h¦  «        }ti          ‰ d         ‰ d         z   ¦  «        D �]í\  Š!\  }}Š"Š ‰" 5                    d ¦  «        r"tm          tT          j+        ‰"d!d…         ¦  «        nR‰" 5                    d"¦  «        r*tm          to          d#¦  «        j8        ‰"dd…         ¦  «        nts          ¦   «         ‰"         Š"|rTtu          ‰"tv          ¦  «        rty          ‰"tT          j+        j=        ¦  «        s t}          t          d$‰"› d%�¦  «        ¦  «        ‚ti          ‰ ¦  «        D ]‡\  }}tu          |t€          ¦  «        rmtƒ          jB        t†          ¦  «        5  |t‰          ¦   «         v rt‰          ¦   «         |         n| E                    |¦  «        ‰ |<   ddd¦  «         n# 1 swxY w Y   Œˆ|d&k    r t�          t�          ||z  ¦  «        d&¦  «        n|x}}‰"|v �ro‰|         ‰ d         }}‰"t"          ur!t‘          t“          ||¦  «        |z  d'¦  «        }‰"tL          u rºt•          ‰ d(         d&k    r:t�          t�          t“          ‰ d(         |d(z  d)z  ¦  «        ¦  «        |z  d&¦  «        n‰ d(         ¦  «        ‰ d(<   t•          |t—          ‰ ¦  «        d*k    r‰ d*         nd+z  ¦  «        }|‰ d(         z  rt‡          d,|› d-|› d.‰ d(         › d/�¦  «        ‚|‰ d&<   ||g‰ d&d…         ¢Š ‰"|v r‰  L                    d(|¦  «         d&}‰"t@          u rd0}|d1v rdg‰ d!d2…<   ‰"tf          u rd0}|d3v r‰  M                    d4¦  «         ‰"td          u rd0}�nK‰"tœ          u r‰|         g‰ ¢Š �n5‰"t!          tž          t           h¦  «        v rJ‰|         ‰ d         ‰ d&         }}}|||g‰ d(d…         ¢Š ‰"t           u r‰  L                    d2|¦  «         d&}�nÎ‰"t¢          u r‰ d!         r‰ d&         n
‰ d&         d2z  }�n¨‰"tT          j+        jR        u r‰|         gŠ �nŠ‰"t¦          u rt©          ˆfd5„|D ¦   «         ¦  «        }�nd‰"t!          tª          t¬          t®          t°          t²          t´          t¶          t¸          tº          t¼          t¾          h¦  «        v r»‰  M                    |
|	ˆfd6„|D ¦   «         g¦  «         ‰"t°          u s‰"t´          u s‰"t²          u s	‰"t¶          u r*t‘          t“          ‰ d(         |¦  «        |z  d'¦  «        ‰ d(<   ‰"tª          t®          t°          t²          t´          t¶          t¸          tº          t¼          t¾          h
v r|‰"_`        �nV‰"tÂ          u rg ‰ dd&…         ¢ˆfd7„|D ¦   «         ‘Š �n1‰"tÄ          u r#‰  c                    ˆfd8„|D ¦   «         ¦  «         �n‰"tÈ          u r"‰  c                    ˆfd9„|D ¦   «         ¦  «         nÚ‰"tÊ          u r#‰  L                    d&ˆfd:„|D ¦   «         ¦  «         n®‰"tÌ          u r#‰  L                    d&ˆfd;„|D ¦   «         ¦  «         n‚‰"tÎ          u r‰ d         }‰|         }||g‰ d&d…         ¢Š nZ‰"tÐ          u r‰|d                  }nB‰"t!          tÒ          tÔ          h¦  «        v r‰ d         }‰|         }g ‰ d&d…         ¢Š n‰|         }|d&k    r-tU          j+        jk        ˆ ˆ"fd<„tÙ          |¦  «        D ¦   «         Ž n ‰"‰ Ž }‰"t.          u r7t—          ‰ ¦  «        d!k    r$|d&k    r|n|gD ]}t          j        |jm        _n        Œt�          ‰"¦  «        d'd=…          o                    d>d¦  «        }t©          d?„ | p                    ¦   «         D ¦   «         ¦  «        |_q        ‰!||c|_r        |_s        |_;        |r,t          j        ‰!d›|d›|d›|jq        d@›d|d›‰ d›�¦  «         | M                    ˆ!fdA„tu          |t”          ¦  «        r|gn|D ¦   «         ¦  «         | c                    |¦  «         ‰!dk    rg Š‰ c                    |¦  «         �ŒïtU          j+        jk        |Ž té          |¦  «        fS )BaO  Parse a YOLO model.yaml dictionary into a PyTorch model.

    Args:
        d (dict): Model dictionary.
        ch (int): Input channels.
        verbose (bool): Whether to print model details.

    Returns:
        (torch.nn.Sequential): PyTorch model.
        (list): Sorted list of layer indices whose outputs need to be saved.
    r   NTÚinfc              3  óB   •K  — | ]}‰                      |¦  «        V — Œd S r  r  ©rƒ   rq   Úds     €rt   rÔ   zparse_model.<locals>.<genexpr>Ü  s-   øè è € Ð\Ð\¨Q §¢ a¡¤Ð\Ð\Ð\Ð\Ð\Ð\rv   )rX  rÖ  Úscalesr{  Úreg_maxé   c              3  óD   •K  — | ]}‰                      |d ¦  «        V — ŒdS )g      ð?Nr  r&  s     €rt   rÔ   zparse_model.<locals>.<genexpr>Þ  s/   øè è € ÐhÐh°˜qŸušu Q¨™}œ}ÐhÐhÐhÐhÐhÐhrv   )Údepth_multipleÚwidth_multiplerÊ  r’  z'no model scale passed. Assuming scale='r  zactivation:rª   ú
rr  z>3Úfromz>20r  r¬   z>10r®   r{  z<45Ú	argumentsz<30r~   rU  r�  znn.rg  ztorchvision.ops.Útorchvisionu   ERROR â�Œï¸� module 'z:' is not a permitted model layer under restricted loading.r‰   é   r§   rq  é   g      à?zC2fAttn hidden channels z
 (from c2=z) must be divisible by nh=z1; adjust width_multiple, nh, or C2fAttn expansionF>   ÚlrŸ   rq   r�  >   r4  rq   )Tg333333ó?c              3  ó(   •K  — | ]}‰|         V — Œd S r  r�   ©rƒ   rq   rý   s     €rt   rÔ   zparse_model.<locals>.<genexpr>d  s'   øè è € Ð&Ð&˜q�R˜”UÐ&Ð&Ð&Ð&Ð&Ð&rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>t  s   ø€ Ð+=Ð+=Ð+=°a¨B¨q¬EÐ+=Ð+=Ð+=rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>z  s   ø€ Ð1Ð1Ð1¨!  1¤Ð1Ð1Ð1rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>|  ó   ø€ Ð*Ð*Ð* 1˜˜AœÐ*Ð*Ð*rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>~  r:  rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>€  ó   ø€ Ð-Ð-Ð- a˜B˜qœEÐ-Ð-Ð-rv   c                ó    •— g | ]
}‰|         ‘ŒS r�   r�   r6  s     €rt   r†   zparse_model.<locals>.<listcomp>‚  r=  rv   c              3  ó"   •K  — | ]	} ‰‰Ž V — Œ
d S r  r�   )rƒ   r¿   rr   rŸ   s     €€rt   rÔ   zparse_model.<locals>.<genexpr>�  s)   øè è € Ð">Ð">° 1 1 d 8Ð">Ð">Ð">Ð">Ð">Ð">rv   rp  z	__main__.c              3  ó>   K  — | ]}|                      ¦   «         V — Œd S r  )Únumelr˜  s     rt   rÔ   zparse_model.<locals>.<genexpr>•  s*   è è € Ð7Ð7 !�A—G’G‘I”IÐ7Ð7Ð7Ð7Ð7Ð7rv   r­   c              3  ó.   •K  — | ]}|d k    ¯|‰z  V — ŒdS )r~   Nr�   )rƒ   rq   r”   s     €rt   rÔ   zparse_model.<locals>.<genexpr>™  s+   øè è € ÐQÐQ˜aÈÈbÊÈ�A˜‘EÈÈÈÈÐQÐQrv   )urÃ  rê   ré   r´   Úiterr7  rG   r¡   r�  r•  rÖ  rt  r%   Údefault_actr¸   rM   r�   r#   r'   r.   r   r-   r   r   r   r   r*   r,   r   r	   r
   r   r   r8   r   r   r   r   r   r   r   r   r–   r   ÚConvTranspose2dr+   r    r6   r   r<   r   r   r*  r�  rÓ   r  ÚopsÚglobalsrl   rº   r¦  r§  rÐ  rN   r	  rõ  ÚsuppressrÙ  ÚlocalsrÁ  rŽ   Úroundr]   rð   r‘   rî   ÚinsertÚextendr   r0   r/   r:   ÚBatchNorm2dr$   r»   r)   rA   rB   r=   r>   rC   rD   r4   r5   r   r   Úlegacyr(   r?   r“   rE   r1   r;   r"   r!   r@   r2   rá  r·   Úcv1rÑ  Úreplacerµ   r¹   r”   r�   Úsorted)#r'  rý   r¦   rÃ  rN  Úmax_channelsrX  rÑ  r(  r{  r)  ÚdepthÚwidthrÊ  r’  r•  Úlayersr•   rø   Úbase_modulesÚrepeat_modulesr�   r  r„   rÂ  Ún_r÷   Úhidden_channelsÚcmÚm_Úblockr¾   rr   r”   rŸ   s#   ``                              @@@rt   ra  ra  Ë  sŽ  øøøøø€ ð €J€J€Jð €FÝ˜‘<”<€LØ\Ð\Ð\Ð\Ð2[Ð\Ñ\Ô\Ñ€BˆˆV�WØ�eŠe�I˜rÑ"Ô"€GØhÐhÐhÐhÐ6gÐhÑhÔhÑ€Eˆ5�)Ø�EŠE�'‰NŒN€EØð 3Øð 	PÝ�˜fŸkšk™mœmÑ,Ô,Ñ-Ô-ˆEÝŒNÐNÀUÐNÐNÐNÑOÔOÐOØ%+¨E¤]Ñ"ˆˆu�lå×%Ò%Ñ'Ô'€JØ
ð <ð 9CÐQ�9×/Ò/°Ñ4Ô4Ð4ÍÈSÉ	Ì	�ÔØð 	<ÝŒK�8 MÑ2Ô2Ð:Ð:°SÐ:Ð:Ñ;Ô;Ð;àð eÝŒÐc˜ÐcÐc ÐcÐc¨CÐcÐc°HÐcÐcÐcÀHÐcÐcÐR]ÐcÐcÐcÑdÔdÐdØ
ˆ€BØ˜2˜r "œv�"ˆD€FÝð#	
Ýð#	
åð#	
õ ð#	
õ ð	#	
õ
 ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ ð#	
õ  ð!#	
õ" ð##	
õ$ ð%#	
õ& ð'#	
õ( ð)#	
õ* ð+#	
õ, ð-#	
õ. ð/#	
õ0 ð1#	
õ2 ð3#	
õ4 ð5#	
õ6 ŒHÔ$ð7#	
õ8 ð9#	
õ: ð;#	
õ< ð=#	
õ> ð?#	
õ@ ðA#	
õB ðC#	
õD ðE#	
ñ%ô %€LõL åÝÝÝÝÝÝÝÝÝÝÝÝÝÝð	
ñô €Nõ& (¨¨*¬¸¸&¼	Ñ(AÑBÔBð rñ rÑˆ‰?ˆAˆq�!�Tð �|Š|˜EÑ"Ô"ð�G•E”H˜a   œeÑ$Ô$Ð$ð �|Š|Ð.Ñ/Ô/ð•� MÑ2Ô2Ô6¸¸"¸#¸#¼Ñ?Ô?Ð?å‘”˜1”ð 	
ð ð 	{�z¨!­TÑ2Ô2ð 	{µzÀ!ÅUÄXÄ_Ñ7UÔ7Uð 	{å�FÐ#x¸1Ð#xÐ#xÐ#xÑyÔyÑzÔzÐzÝ˜d‘O”Oð 	Tð 	T‰DˆAˆqÝ˜!�SÑ!Ô!ð TÝÔ(­Ñ4Ô4ð Tð TØ-.µ&±(´(¨]¨]�f™hœh qœk˜kÀ×@PÒ@PÐQRÑ@SÔ@S�D˜‘GðTð Tð Tñ Tô Tð Tð Tð Tð Tð Tð Tøøøð Tð Tð Tð Tøà-.°ªU¨U••U˜1˜u™9Ñ%Ô% qÑ)Ô)Ð)¸Ð9ˆˆBØ�ÐÑØ˜”U˜D œG�ˆBØ�Ð Ð Ý#¥C¨¨LÑ$9Ô$9¸EÑ$AÀ1ÑEÔE�Ø•Gˆ|ˆ|ÝÐ^bÐcdÔ^eÐhiÒ^iÐ^i�c¥%­¨D°¬G°\ÀQÑ5FÈ"Ñ5LÑ(MÔ(MÑ"NÔ"NÐQVÑ"VÐXYÑZÔZÐZÐosÐtuÔovÑwÔw��Q‘Ý"% bµs¸4±y´yÀ1²}°}¨D°¬G¨GÈ#Ñ&NÑ"OÔ"O�Ø" T¨!¤WÑ,ð Ý$ðJ°?ð Jð JÈbð Jð JÐlpÐqrÔlsð Jð Jð Jñô ð ð *��Q‘à˜Ð&˜T ! " "œXÐ&ˆDØ�NÐ"Ð"Ø—’˜A˜qÑ!Ô!Ð!Ø�Ø•DˆyˆyØ�Ø˜OÐ+Ð+Ø!% �D˜˜1˜‘IØ•EˆzˆzØ�Ø˜JÐ&Ð&Ø—K’K Ñ,Ô,Ð,Ø•Fˆ{ˆ{Ø�ùØ•$ˆYˆYØ�q”E�>˜D�>ˆD‰DØ•)�V¥WÐ-Ñ.Ô.Ð.Ð.Ø˜Aœ  Q¤¨¨a¬�B�ˆBØ˜˜BÐ*  a b b¤Ð*ˆDØ•Gˆ|ˆ|Ø—’˜A˜qÑ!Ô!Ð!Ø�ùØ•+ÐÐØ  œGÐ4��a”�¨¨a¬°1©ˆB‰BØ•%”(Ô&Ð&Ð&Ø�q”E�7ˆD‰DØ•&ˆ[ˆ[ÝÐ&Ð&Ð&Ð& AÐ&Ñ&Ô&Ñ&Ô&ˆB‰BØ•)åÝÝÝÝÝÝÝÝÝÝðñ
ô 
ð 
ð 
ð �KŠK˜ 'Ð+=Ð+=Ð+=Ð+=¸1Ð+=Ñ+=Ô+=Ð>Ñ?Ô?Ð?Ø•Gˆ|ˆ|˜q¥LÐ0Ð0°A½°N°NÀaÍ>ÐFYÐFYÝ(­¨T°!¬W°lÑ)CÔ)CÀeÑ)KÈQÑOÔO��Q‘Ø•V�[­'µ9½lÍNÕ\`ÕbhÕjmÕotÐuÐuÐuØ!�”ùØ•%ˆZˆZØ2�T˜"˜1˜"”XÐ2Ð1Ð1Ð1Ð1¨qÐ1Ñ1Ô1Ð2ˆD‰DØ•/Ð!Ð!Ø�KŠKÐ*Ð*Ð*Ð*¨Ð*Ñ*Ô*Ñ+Ô+Ð+Ñ+Ø•)ˆ^ˆ^Ø�KŠKÐ*Ð*Ð*Ð*¨Ð*Ñ*Ô*Ñ+Ô+Ð+Ð+ØÕ"Ð"Ð"Ø�KŠK˜Ð-Ð-Ð-Ð-¨1Ð-Ñ-Ô-Ñ.Ô.Ð.Ð.Ø•-ÐÐØ�KŠK˜Ð-Ð-Ð-Ð-¨1Ð-Ñ-Ô-Ñ.Ô.Ð.Ð.Ø•(ˆ]ˆ]Ø�a”ˆBØ�A”ˆBØ˜Ð&˜T ! " "œXÐ&ˆDˆDØ•&ˆ[ˆ[Ø�A�b”E”ˆBˆBØ•)�[­%Ð0Ñ1Ô1Ð1Ð1Ø�a”ˆBØ�A”ˆBØ�T˜!˜"˜"”X�;ˆDˆDà�A”ˆBàCDÀqÂ5À5�UŒXÔ Ð">Ð">Ð">Ð">Ð">µU¸1±X´XÐ">Ñ">Ô">Ð?Ð?ÈaÈaÐQUÈhˆØ•ˆ9ˆ9�˜T™œ aš˜Ø  1šu˜u˜˜¨2¨$ð 1ð 1�Ý $Ô 0�”	”�Ý�‰FŒF�1�R�4ŒL× Ò  ¨bÑ1Ô1ˆÝÐ7Ð7 r§}¢}¡¤Ð7Ñ7Ô7Ñ7Ô7ˆŒØ  AÐˆŒˆbŒd�B”GØð 	UÝŒK˜1ÐSÐS ÐSÐS¨"ÐSÐS°´ÐSÐSÐSÀÐSÐSÀtÐSÐSÐSÑTÔTÐTØ�ŠÐQÐQÐQÐQ­:°a½Ñ+=Ô+=Ð$D Q C CÀ1ÐQÑQÔQÑQÔQÐQØ�Š�bÑÔÐØ�Š6ˆ6ØˆBØ
�	Š	�"‰Œˆ‰ÝŒ8Ô Ð'­°©¬Ð5Ð5s   Ð-=Q6Ñ6Q:Ñ=Q:c                óð  — t          | ¦  «        } | j        d„ dD ¦   «         v rXt          j        dd| j        ¦  «        }t	          j        d| j        › d|› d�¦  «         |                      || j        z   ¦  «        } t          j        dd	t          | ¦  «        ¦  «        }t          | d
¬¦  «        pt          |¦  «        }t          j        |¦  «        }t          | ¦  «        |d<   t          | ¦  «        |d<   |S )z•Load a YOLO model from a YAML file.

    Args:
        path (str | Path): Path to the YAML file.

    Returns:
        (dict): Model dictionary.
    c              3  ó0   K  — | ]}d D ]}d|› |› d�V — ŒŒdS ))é   r2  ÚyolovÚ6Nr�   r&  s      rt   rÔ   z"yaml_model_load.<locals>.<genexpr>«  s;   è è € ÐFÐF¨¸vÐFÐF¸!�_˜Q�_ �_�_�_ÐFÐFÐFÐFÐFÐFÐFrv   Únsmlxz(\d+)([nslmx])6(.+)?$z	\1\2-p6\3z8Ultralytics YOLO P6 models now use -p6 suffix. Renaming z to r)  z(\d+)([nslmx])(.+)?$z\1\3F)Úhardr’  Ú	yaml_file)r   ÚstemÚreÚsubrG   r¡   Ú	with_namerû  r	  rS   rK   rÿ   Úguess_model_scale)ÚpathÚnew_stemÚunified_pathrd  r'  s        rt   r_  r_  ¡  sï   € õ �‰:Œ:€DØ„yÐFÐF¨gÐFÑFÔFÐFÐFÝ”6Ð2°LÀ$Ä)ÑLÔLˆÝŒÐlÐRVÔR[ÐlÐlÐaiÐlÐlÐlÑmÔmÐmØ�~Š~˜h¨¬Ñ4Ñ5Ô5ˆå”6Ð1°7½CÀ¹I¼IÑFÔF€LÝ˜4 eÐ,Ñ,Ô,ÐHµ
¸<Ñ0HÔ0H€IÝŒ	�)ÑÔ€AÝ" 4Ñ(Ô(€A€g�JÝ˜‘Y”Y€A€k�NØ€Hrv   c                óš   — 	 t          j        dt          | ¦  «        j        ¦  «                             d¦  «        S # t
          $ r Y dS w xY w)a%  Extract the size character n, s, m, l, or x of the model's scale from the model path.

    Args:
        model_path (str | Path): The path to the YOLO model's YAML file.

    Returns:
        (str): The size character of the model's scale (n, s, m, l, or x), or empty string if not found.
    zyolo(e-)?[v]?\d+([nslmx])r§   rr  )rf  Úsearchr   re  ÚgroupÚAttributeError)Ú
model_paths    rt   ri  ri  ¸  sV   € ðÝŒyÐ5µt¸JÑ7GÔ7GÔ7LÑMÔM×SÒSÐTUÑVÔVÐVøÝð ð ð Øˆrˆrðøøøs   ‚9< ¼
A
Á	A
c                óÂ  — d„ }t          | t          ¦  «        r<t          j        t          ¦  «        5   || ¦  «        cddd¦  «         S # 1 swxY w Y   t          | t
          j        j        ¦  «        �rvdD ]J}t          j        t          ¦  «        5  t          |¦  «        d         cddd¦  «         c S # 1 swxY w Y   ŒKdD ]M}t          j        t          ¦  «        5   |t          |¦  «        ¦  «        cddd¦  «         c S # 1 swxY w Y   ŒN|  	                    ¦   «         D ]Ä}t          |t          ¦  «        r dS t          |t          t          f¦  «        r dS t          |t          ¦  «        r dS t          |t          ¦  «        r d	S t          |t          ¦  «        r d
S t          |t           ¦  «        r dS t          |t"          t$          t&          t(          f¦  «        r dS ŒÅt          | t*          t,          f¦  «        rÄddlm} |                     | ¦  «                             d¦  «        x}r|S t-          | ¦  «        } d| j        v s	d| j        v rdS d| j        v s	d| j        v rdS d| j        v s	d| j        v rdS d| j        v s	d	| j        v rd	S d| j        v s	d
| j        v rd
S d| j        v s	d| j        v rdS d| j        v rdS t;          j        d¦  «         dS )aA  Guess the task of a PyTorch model from its architecture or configuration.

    Args:
        model (torch.nn.Module | dict | str | Path): PyTorch model, model configuration dict, or model file path.

    Returns:
        (str): Task of the model ('detect', 'segment', 'semantic', 'depth', 'classify', 'pose', 'obb').
    c                ó¦   — | d         d         d                               ¦   «         }|dv rdS d|v rdS d|v rdS d	|v rd	S d
|v rd
S d|v rdS d|v rdS dS )zGuess from YAML dictionary.r�  r~   rp  >   Úfcr”  ÚclassifyÚ
classifierru  ÚdetectÚsemanticsegmentÚsemanticÚsegmentÚposeÚobbrS  N)r  )rb  rŸ   s     rt   Úcfg2taskz"guess_model_task.<locals>.cfg2taskÑ  sŽ   € à�ŒK˜ŒO˜BÔ×%Ò%Ñ'Ô'ˆØÐ7Ð7Ð7Ø�:Ø�qˆ=ˆ=Ø�8Ø Ð!Ð!Ø�:Ø˜ˆ>ˆ>Ø�9Ø�Qˆ;ˆ;Ø�6Ø�Aˆ:ˆ:Ø�5Ø�aˆ<ˆ<Ø�7ð ˆ<rv   N)z
model.argszmodel.model.argszmodel.model.model.argsr  )z
model.yamlzmodel.model.yamlzmodel.model.model.yamlry  rz  ru  r{  r|  rS  rw  r   )ÚBaseBackendz-semz-segz-clsz-posez-obbz-depthz¹Unable to automatically guess model task, assuming 'task=detect'. Explicitly define task for your model, i.e. 'task=detect', 'segment', 'semantic', 'depth', 'classify', 'pose' or 'obb'.)rl   rm   rõ  rH  rÏ  r–   r   r§  rt  rÅ   r?   r=   rC   r#   r4   r   r(   r)   rA   rB   rE   r	  r   Úultralytics.nn.backends.baser~  Úread_metadataré   re  ÚpartsrG   r¡   )r�   r}  rq   rŸ   r~  r  s         rt   r  r  Ç  s  € ðð ð õ& �%�ÑÔð #ÝÔ ¥Ñ+Ô+ð 	#ð 	#Ø�8˜E‘?”?ð	#ð 	#ð 	#ð 	#ñ 	#ô 	#ð 	#ð 	#ð 	#ð 	#ð 	#ð 	#øøøð 	#ð 	#ð 	#ð 	#õ �%�œœÑ)Ô)ñ  ØKð 	'ð 	'ˆAÝÔ$¥YÑ/Ô/ð 'ð 'Ý˜A‘w”w˜v”ð'ð 'ð 'ð 'ñ 'ô 'ð 'ð 'ð 'ð 'ð 'ð 'ð 'ð 'øøøð 'ð 'ð 'ð 'ð 'àKð 	)ð 	)ˆAÝÔ$¥YÑ/Ô/ð )ð )Ø�x¥ Q¡¤Ñ(Ô(ð)ð )ð )ð )ñ )ô )ð )ð )ð )ð )ð )ð )ð )ð )øøøð )ð )ð )ð )ð )à—’‘”ð 	 ð 	 ˆAÝ˜!�_Ñ-Ô-ð  Ø!�z�zÝ˜A¥­Ð6Ñ7Ô7ð  Ø �y�yÝ˜A�xÑ(Ô(ð 	 Ø!�z�zÝ˜A�tÑ$Ô$ð  Ø�v�vÝ˜A�sÑ#Ô#ð  Ø�u�uÝ˜A�uÑ%Ô%ð  Ø�w�wÝ˜A¥­µ[Å)ÐLÑMÔMð  Ø�x�xð õ �%�#�t˜Ñ%Ô%ð Ø<Ð<Ð<Ð<Ð<Ð<à×,Ò,¨UÑ3Ô3×7Ò7¸Ñ?Ô?Ð?ˆ4ð 	ØˆKõ �U‘”ˆØ�U”ZÐÐ :°´Ð#<Ð#<Ø�:Ø�u”zÐ!Ð! Y°%´+Ð%=Ð%=Ø�9Ø�u”zÐ!Ð! Z°5´;Ð%>Ð%>Ø�:Ø˜œ
Ð"Ð" f°´Ð&;Ð&;Ø�6Ø�u”zÐ!Ð! U¨e¬kÐ%9Ð%9Ø�5Ø˜œÐ#Ð# w°%´+Ð'=Ð'=Ø�7Ø˜œÐ$Ð$Ø�8õ „Nð	ñô ð ð
 ˆ8s5   ²A
Á
AÁAÂB6Â6B:	Â=B:	ÃDÄD
	ÄD
	rÈ  r  )NTFrM  )”Ú
__future__r   rõ  r  rÜ  rf  rñ  r³   r   rç  r   r  r¹   r–   r   Úultralytics.nn.autobackendr   Úultralytics.nn.modulesr   r	   r
   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r    r!   r"   r#   r$   r%   r&   r'   r(   r)   r*   r+   r,   r-   r.   r/   r0   r1   r2   r3   r4   r5   r6   r7   r8   r9   r:   r;   r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   Úultralytics.utilsrF   rG   rH   rI   rJ   rK   rL   rM   rN   Úultralytics.utils.checksrO   rP   rQ   rR   rS   rn  rT   rU   rV   rW   rX   rY   rZ   r[   r\   Úultralytics.utils.opsr]   Úultralytics.utils.patchesr^   Úultralytics.utils.torch_utilsr_   r`   ra   rb   rc   rd   re   rf   rg   rh   r§  rj   rc  re  r£  r­  r²  rÆ  rÎ  rÓ  rê  r  r<  rq  rV  rw  ÚRLockr�  rö  r�  r�  r  r"  ra  r_  ri  r  r�   rv   rt   ú<module>r‹     s`  ðð #Ð "Ð "Ð "Ð "Ð "à Ð Ð Ð Ø Ð Ð Ð Ø €€€Ø 	€	€	€	Ø Ð Ð Ð Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð à Ð Ð Ð Ø €€€Ø Ð Ð Ð Ð Ð à 8Ð 8Ð 8Ð 8Ð 8Ð 8ð?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?ð@
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð tÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sÐ sð
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 
ð 1Ð 0Ð 0Ð 0Ð 0Ð 0Ø 0Ð 0Ð 0Ð 0Ð 0Ð 0ðð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ðuZð uZð uZð uZð uZ�””ñ uZô uZð uZðp	4ð 4ð 4ð&^tð ^tð ^tð ^tð ^t�Yñ ^tô ^tð ^tðByð yð yð yð yˆ~ñ yô yð yð@!
ð !
ð !
ð !
ð !
˜ñ !
ô !
ð !
ðHEð Eð Eð Eð E 	ñ Eô Eð EðP'oð 'oð 'oð 'oð 'o�ñ 'oô 'oð 'oðT!ð !ð !ð !ð !�ñ !ô !ð !ð(Y&ð Y&ð Y&ð Y&ð Y&˜)ñ Y&ô Y&ð Y&ðxgð gð gð gð g˜>ñ gô gð gðT|,ð |,ð |,ð |,ð |,�ñ |,ô |,ð |,ð~G,ð G,ð G,ð G,ð G,�ñ G,ô G,ð G,ðT0*ð 0*ð 0*ð 0*ð 0*�JÐ 1ñ 0*ô 0*ð 0*ðf&ð &ð &ð &ð &ˆuŒxÔ"ñ &ô &ð &ðX *˜)œ/Ñ+Ô+Ð ð Ôð61ð 61ð 61ñ Ôð61ðrl
ð l
ð l
ð l
ð l
ñ l
ô l
ð l
ð^Nð Nð Nð Nðb.ð .ð .ð .ðbS6ð S6ð S6ð S6ðlð ð ð.ð ð ðUð Uð Uð Uð Urv   