§
    9²¸j¼Ï  ã                   ó¤  — d 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m	Z	 ddl
mZ ddlmZ dd	lmZ dd
lmZ ddlmZmZmZmZ ddlmZ ddlmZ ddlmZmZmZ ddl m!Z!  ej"        e#¦  «        Z$ ed¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Z% G d„ de	j&        ¦  «        Z' G d„ de	j&        ¦  «        Z( G d„ de	j&        ¦  «        Z) G d„ de)¦  «        Z* G d„ de	j&        ¦  «        Z+e)e*dœZ, G d „ d!e	j&        ¦  «        Z- G d"„ d#e	j&        ¦  «        Z. G d$„ d%e	j&        ¦  «        Z/ G d&„ d'e	j&        ¦  «        Z0 G d(„ d)e¦  «        Z1 G d*„ d+e	j&        ¦  «        Z2 G d,„ d-e	j&        ¦  «        Z3e G d.„ d/e¦  «        ¦   «         Z4e G d0„ d1e4¦  «        ¦   «         Z5 G d2„ d3e	j&        ¦  «        Z6 ed4¬¦  «         G d5„ d6e4¦  «        ¦   «         Z7 G d7„ d8e	j&        ¦  «        Z8 G d9„ d:e	j&        ¦  «        Z9 G d;„ d<e	j&        ¦  «        Z: G d=„ d>e	j&        ¦  «        Z; G d?„ d@e	j&        ¦  «        Z<e G dA„ dBe4¦  «        ¦   «         Z=g dC¢Z>dS )DzPyTorch Data2VecVision model.é    N)Ú	dataclass)ÚOptional)Únn)ÚCrossEntropyLossé   )Úinitialization)ÚACT2FN)ÚGradientCheckpointingLayer)ÚBaseModelOutputÚBaseModelOutputWithPoolingÚImageClassifierOutputÚSemanticSegmenterOutput)ÚPreTrainedModel)Ú#compile_compatible_method_lru_cache)Úauto_docstringÚloggingÚ	torch_inté   )ÚData2VecVisionConfigz7
    Class for outputs of [`Data2VecVisionModel`].
    )Úcustom_introc                   ó   — e Zd ZdZdS )Ú$Data2VecVisionModelOutputWithPoolingaF  
    pooler_output (`torch.FloatTensor` of shape `(batch_size, hidden_size)`):
        Average of the last layer hidden states of the patch tokens (excluding the *[CLS]* token) if
        *config.use_mean_pooling* is set to True. If set to False, then the final hidden state of the *[CLS]* token
        will be returned.
    N)Ú__name__Ú
__module__Ú__qualname__Ú__doc__© ó    ú…/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/data2vec/modeling_data2vec_vision.pyr   r   +   s   € € € € € ðð ð ð r   r   c                   ó”   ‡ — e Zd ZdZdeddfˆ fd„Zdej        dededej        fd	„Z		 dd
ej        dej
        dz  dej        fd„Zˆ xZS )ÚData2VecVisionEmbeddingszc
    Construct the CLS token, position and patch embeddings. Optionally, also the mask token.

    ÚconfigÚreturnNc                 óÆ  •— t          ¦   «                              ¦   «          t          j        t	          j        dd|j        ¦  «        ¦  «        | _        |j        r3t          j        t	          j        dd|j        ¦  «        ¦  «        | _	        nd | _	        t          |¦  «        | _        |j        | _        t          |j        t          j        j        ¦  «        r|j        n|j        |j        f| _        | j        j        }|j        r6t          j        t	          j        d|dz   |j        ¦  «        ¦  «        | _        nd | _        t          j        |j        ¦  «        | _        d S )Nr   )ÚsuperÚ__init__r   Ú	ParameterÚtorchÚzerosÚhidden_sizeÚ	cls_tokenÚuse_mask_tokenÚ
mask_tokenÚData2VecVisionPatchEmbeddingsÚpatch_embeddingsÚ
patch_sizeÚ
isinstanceÚ
image_sizeÚcollectionsÚabcÚIterableÚnum_patchesÚ use_absolute_position_embeddingsÚposition_embeddingsÚDropoutÚhidden_dropout_probÚdropout)Úselfr"   r6   Ú	__class__s      €r   r&   z!Data2VecVisionEmbeddings.__init__B   s(  ø€ Ý‰Œ×ÒÑÔÐåœ¥e¤k°!°Q¸Ô8JÑ&KÔ&KÑLÔLˆŒØÔ ð 	#Ý œl­5¬;°q¸!¸VÔ=OÑ+PÔ+PÑQÔQˆDŒOˆOà"ˆDŒOÝ =¸fÑ EÔ EˆÔØ Ô+ˆŒõ ˜&Ô+­[¬_Ô-EÑFÔFð8ˆFÔÐàÔ# VÔ%6Ð7ð 	Œð
 Ô+Ô7ˆØÔ2ð 	,Ý')¤|µE´KÀÀ;ÐQRÁ?ÐTZÔTfÑ4gÔ4gÑ'hÔ'hˆDÔ$Ð$à'+ˆDÔ$Ý”z &Ô"<Ñ=Ô=ˆŒˆˆr   Ú
embeddingsÚheightÚwidthc                 ó”  — |j         d         dz
  }| j        j         d         dz
  }t          j                             ¦   «         s||k    r||k    r| j        S | j        dd…dd…f         }| j        dd…dd…f         }|j         d         }|| j        z  }	|| j        z  }
t          |dz  ¦  «        }|                     d|||¦  «        }|                     dddd¦  «        }t          j
                             ||	|
fdd	¬
¦  «        }|                     dddd¦  «                             dd|¦  «        }t          j        ||fd¬¦  «        S )a   
        This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution
        images. This method is also adapted to support torch.jit tracing.

        Adapted from:
        - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and
        - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211
        r   Néÿÿÿÿg      à?r   r   é   ÚbicubicF©ÚsizeÚmodeÚalign_corners©Údim)Úshaper8   r(   ÚjitÚ
is_tracingr0   r   ÚreshapeÚpermuter   Ú
functionalÚinterpolateÚviewÚcat)r<   r>   r?   r@   r6   Únum_positionsÚclass_pos_embedÚpatch_pos_embedrJ   Ú
new_heightÚ	new_widthÚsqrt_num_positionss               r   Úinterpolate_pos_encodingz1Data2VecVisionEmbeddings.interpolate_pos_encodingY   sr  € ð !Ô& qÔ)¨AÑ-ˆØÔ0Ô6°qÔ9¸AÑ=ˆõ Œy×#Ò#Ñ%Ô%ð 	,¨+¸Ò*FÐ*FÈ6ÐUZÊ?È?ØÔ+Ð+àÔ2°1°1°1°b°q°b°5Ô9ˆØÔ2°1°1°1°a°b°b°5Ô9ˆàÔ˜rÔ"ˆà˜tœÑ.ˆ
Ø˜Tœ_Ñ,ˆ	å& }°cÑ'9Ñ:Ô:ÐØ)×1Ò1°!Ð5GÐI[Ð]`ÑaÔaˆØ)×1Ò1°!°Q¸¸1Ñ=Ô=ˆåœ-×3Ò3ØØ˜iÐ(ØØð	 4ñ 
ô 
ˆð *×1Ò1°!°Q¸¸1Ñ=Ô=×BÒBÀ1ÀbÈ#ÑNÔNˆåŒy˜/¨?Ð;ÀÐCÑCÔCÐCr   Úpixel_valuesÚbool_masked_posc                 ó  — |j         \  }}}}|                      |¦  «        \  }\  }}|                     ¦   «         \  }	}
}|�R| j                             |	|
d¦  «        }|                     d¦  «                             |¦  «        }|d|z
  z  ||z  z   }| j                             |	dd¦  «        }t          j	        ||fd¬¦  «        }| j
        �||                      |||¦  «        z   }|                      |¦  «        }|||ffS ©NrB   r   rI   )rK   r/   rF   r-   ÚexpandÚ	unsqueezeÚtype_asr+   r(   rS   r8   rZ   r;   )r<   r[   r\   Ú_r?   r@   r>   Úpatch_heightÚpatch_widthÚ
batch_sizeÚseq_lenÚmask_tokensÚwÚ
cls_tokenss                 r   Úforwardz Data2VecVisionEmbeddings.forward�   s  € ð
 +Ô0Ñˆˆ1ˆf�eØ26×2GÒ2GÈÑ2UÔ2UÑ/ˆ
Ñ/�\ ;Ø!+§¢Ñ!2Ô!2Ñˆ
�G˜QàÐ&Øœ/×0Ò0°¸WÀbÑIÔIˆKà×)Ò)¨"Ñ-Ô-×5Ò5°kÑBÔBˆAØ# q¨1¡uÑ-°¸a±Ñ?ˆJà”^×*Ò*¨:°r¸2Ñ>Ô>ˆ
Ý”Y 
¨JÐ7¸QÐ?Ñ?Ô?ˆ
àÔ#Ð/Ø# d×&CÒ&CÀJÐPVÐX]Ñ&^Ô&^Ñ^ˆJà—\’\ *Ñ-Ô-ˆ
à˜L¨+Ð6Ð6Ð6r   ©N)r   r   r   r   r   r&   r(   ÚTensorÚintrZ   Ú
BoolTensorrj   Ú__classcell__©r=   s   @r   r!   r!   <   sÜ   ø€ € € € € ðð ð
>Ð3ð >¸ð >ð >ð >ð >ð >ð >ð.&D°5´<ð &DÈð &DÐUXð &DÐ]bÔ]ið &Dð &Dð &Dð &DðV 48ð7ð 7à”lð7ð Ô)¨DÑ0ð7ð 
Œð	7ð 7ð 7ð 7ð 7ð 7ð 7ð 7r   r!   c                   óF   ‡ — e Zd ZdZˆ fd„Zdej        dej        fd„Zˆ xZS )r.   zì
    This class turns `pixel_values` of shape `(batch_size, num_channels, height, width)` into the initial
    `hidden_states` (patch embeddings) of shape `(batch_size, seq_length, hidden_size)` to be consumed by a
    Transformer.
    c                 ó  •— t          ¦   «                              ¦   «          |j        |j        }}|j        |j        }}t          |t          j        j	        ¦  «        r|n||f}t          |t          j        j	        ¦  «        r|n||f}|d         |d         z  |d         |d         z  z  }|d         |d         z  |d         |d         z  f}|| _        || _        || _        || _
        || _        t          j        ||||¬¦  «        | _        d S )Nr   r   ©Úkernel_sizeÚstride)r%   r&   r2   r0   Únum_channelsr*   r1   r3   r4   r5   r6   Úpatch_shaper   ÚConv2dÚ
projection)	r<   r"   r2   r0   rv   r*   r6   rw   r=   s	           €r   r&   z&Data2VecVisionPatchEmbeddings.__init__£   s  ø€ Ý‰Œ×ÒÑÔÐØ!'Ô!2°FÔ4E�Jˆ
Ø$*Ô$7¸Ô9K�kˆå#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ý#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ø! !”}¨
°1¬Ñ5¸*ÀQ¼-È:ÐVWÌ=Ñ:XÑYˆØ! !”}¨
°1¬Ñ5°zÀ!´}È
ÐSTÌÑ7UÐVˆØ$ˆŒØ$ˆŒØ(ˆÔØ&ˆÔØ&ˆÔåœ) L°+È:Ð^hÐiÑiÔiˆŒˆˆr   r[   r#   c                 óN  — |j         \  }}}}|| j        k    rt          d¦  «        ‚|                      |                     | j        j        j        ¦  «        ¦  «        }|j         d         |j         d         }}|                     d¦  «                             dd¦  «        }|||ffS )NzeMake sure that the channel dimension of the pixel values match with the one set in the configuration.rC   r   r   )	rK   rv   Ú
ValueErrorry   ÚtoÚweightÚdtypeÚflattenÚ	transpose)	r<   r[   re   rv   r?   r@   r>   rc   rd   s	            r   rj   z%Data2VecVisionPatchEmbeddings.forward´   s©   € Ø2>Ô2DÑ/ˆ
�L &¨%Ø˜4Ô,Ò,Ð,ÝØwñô ð ð —_’_ \§_¢_°T´_Ô5KÔ5QÑ%RÔ%RÑSÔSˆ
Ø$.Ô$4°QÔ$7¸Ô9IÈ!Ô9L�kˆØ×'Ò'¨Ñ*Ô*×4Ò4°Q¸Ñ:Ô:ˆ
à˜L¨+Ð6Ð6Ð6r   )	r   r   r   r   r&   r(   rl   rj   ro   rp   s   @r   r.   r.   œ   sm   ø€ € € € € ðð ðjð jð jð jð jð"7 E¤Lð 7°U´\ð 7ð 7ð 7ð 7ð 7ð 7ð 7ð 7r   r.   c                   óÊ   ‡ — e Zd Zddededz  ddfˆ fd„Z	 	 	 	 ddej        ded	ej        dz  d
edee	         dz  deej                 eej        ej        f         z  fd„Z
ˆ xZS )ÚData2VecVisionSelfAttentionNr"   Úwindow_sizer#   c                 óÜ  •— t          ¦   «                              ¦   «          || _        |j        |j        z  dk    r0t          |d¦  «        s t          d|j        › d|j        › d�¦  «        ‚|j        | _        t          |j        |j        z  ¦  «        | _        | j        | j        z  | _	        t          j        |j        | j	        ¦  «        | _        t          j        |j        | j	        d¬¦  «        | _        t          j        |j        | j	        ¦  «        | _        t          j        |j        ¦  «        | _        t%          |¦  «        | _        | j        rt)          ||¬¦  «        | _        d S d S )	Nr   Úembedding_sizezThe hidden size z4 is not a multiple of the number of attention heads ú.F)Úbias©rƒ   )r%   r&   r"   r*   Únum_attention_headsÚhasattrr{   rm   Úattention_head_sizeÚall_head_sizer   ÚLinearÚqueryÚkeyÚvaluer9   Úattention_probs_dropout_probr;   ÚboolÚhas_relative_position_biasÚ"Data2VecVisionRelativePositionBiasÚrelative_position_bias©r<   r"   rƒ   r=   s      €r   r&   z$Data2VecVisionSelfAttention.__init__Ä   se  ø€ Ý‰Œ×ÒÑÔÐØˆŒØÔ Ô :Ñ:¸aÒ?Ð?ÍÐPVÐXhÑHiÔHiÐ?Ýð7 6Ô#5ð 7ð 7ØÔ3ð7ð 7ð 7ñô ð ð
 $*Ô#=ˆÔ Ý#& vÔ'9¸FÔ<VÑ'VÑ#WÔ#WˆÔ Ø!Ô5¸Ô8PÑPˆÔå”Y˜vÔ1°4Ô3EÑFÔFˆŒ
Ý”9˜VÔ/°Ô1CÈ%ÐPÑPÔPˆŒÝ”Y˜vÔ1°4Ô3EÑFÔFˆŒ
å”z &Ô"EÑFÔFˆŒå*.¨{Ñ*;Ô*;ˆÔ'ØÔ*ð 	nÝ*LÈVÐalÐ*mÑ*mÔ*mˆDÔ'Ð'Ð'ð	nð 	nr   FÚhidden_statesÚoutput_attentionsr•   rZ   Ú
resolutionc                 ó:  — |j         d d…         }g |¢d‘| j        ‘R }|                      |¦  «                             |¦  «                             dd¦  «        }|                      |¦  «                             |¦  «                             dd¦  «        }	|                      |¦  «                             |¦  «                             dd¦  «        }
t          j        ||	                     dd¦  «        ¦  «        }|t          j
        | j        ¦  «        z  }| j        rI|\  }}|| j        j        z  || j        j        z  f}||                      |||j         d         ¬¦  «        z   }|�||z   }t          j                             |d¬¦  «        }|                      |¦  «        }t          j        ||
¦  «        }|                     dddd¦  «                             ¦   «         }|                     ¦   «         d d…         | j        fz   } |j        |Ž }|r||fn|f}|S )	NrB   r   rC   éþÿÿÿ©Údim_sizerI   r   r   )rK   r‹   rŽ   rR   r€   r�   r�   r(   ÚmatmulÚmathÚsqrtr“   r"   r0   r•   r   rP   Úsoftmaxr;   rO   Ú
contiguousrF   rŒ   )r<   r—   r˜   r•   rZ   r™   Úinput_shapeÚhidden_shapeÚquery_layerÚ	key_layerÚvalue_layerÚattention_scoresr?   r@   rƒ   Úattention_probsÚcontext_layerÚnew_context_layer_shapeÚoutputss                      r   rj   z#Data2VecVisionSelfAttention.forwardÛ   s.  € ð $Ô)¨#¨2¨#Ô.ˆØC˜ÐC bÐC¨$Ô*BÐCÐCˆØ—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆØ—H’H˜]Ñ+Ô+×0Ò0°Ñ>Ô>×HÒHÈÈAÑNÔNˆ	Ø—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆõ !œ<¨°Y×5HÒ5HÈÈRÑ5PÔ5PÑQÔQÐà+­d¬i¸Ô8PÑ.QÔ.QÑQÐð Ô*ð 	Ø&‰MˆF�EØ! T¤[Ô%;Ñ;¸UÀdÄkÔF\Ñ=\Ð]ˆKØ/°$×2MÒ2MØÐ5ÀÔ@SÐTUÔ@Vð 3Nñ 3ô 3ñ  Ðð
 "Ð-Ø/Ð2HÑHÐõ œ-×/Ò/Ð0@ÀbÐ/ÑIÔIˆð Ÿ,š, Ñ7Ô7ˆåœ _°kÑBÔBˆà%×-Ò-¨a°°A°qÑ9Ô9×DÒDÑFÔFˆØ"/×"4Ò"4Ñ"6Ô"6°s¸°sÔ";¸tÔ?QÐ>SÑ"SÐØ*˜Ô*Ð,CÐDˆà6GÐ]�= /Ð2Ð2ÈmÐM]ˆàˆr   rk   ©FNFN)r   r   r   r   Útupler&   r(   rl   r’   rm   rj   ro   rp   s   @r   r‚   r‚   Ã   sô   ø€ € € € € ðnð nÐ3ð nÀ%È$Á,ð nÐZ^ð nð nð nð nð nð nð4 #(Ø6:Ø).Ø(,ð.ð .à”|ð.ð  ð.ð !&¤¨tÑ 3ð	.ð
 #'ð.ð ˜#”J Ñ%ð.ð 
ˆuŒ|Ô	˜u U¤\°5´<Ð%?Ô@Ñ	@ð.ð .ð .ð .ð .ð .ð .ð .r   r‚   c                   ó¤   — e Zd Z	 	 	 	 d
dej        dedej        dz  dedee         dz  deej                 eej        ej        f         z  fd	„ZdS )ÚData2VecVisionSdpaSelfAttentionFNr—   r˜   r•   rZ   r™   r#   c           	      ó  — |r't                                | j        j        › d�¦  «         |j        d d…         }g |¢d‘| j        ‘R }|                      |¦  «                             |¦  «                             dd¦  «        }|  	                    |¦  «                             |¦  «                             dd¦  «        }	|  
                    |¦  «                             |¦  «                             dd¦  «        }
d }| j        rF|\  }}|| j        j        z  || j        j        z  f}|                      |||j        d         ¬¦  «        }|�
|€|}n||z  }dt          j        | j        ¦  «        z  }t"          j        j                             ||	|
|| j        r| j        j        ndd|¬¦  «        }|                     d	ddd
¦  «                             ¦   «         }|                     ¦   «         d d…         | j        fz   } |j        |Ž }|d fS )NzÁ does not support `output_attentions=True`. The returned attention weights will be `None`. If you want to get attention weights, please set `attn_implementation='eager'` when loading the model.rB   r   rC   rœ   ç        F)Ú	attn_maskÚ	dropout_pÚ	is_causalÚscaler   r   r›   )ÚloggerÚwarning_oncer=   r   rK   r‹   rŽ   rR   r€   r�   r�   r“   r"   r0   r•   rŸ   r    r(   r   rP   Úscaled_dot_product_attentionÚtrainingr‘   rO   r¢   rF   rŒ   )r<   r—   r˜   r•   rZ   r™   r£   r¤   r¥   r¦   r§   Ú	attn_biasr?   r@   rƒ   Úscalingrª   r«   s                     r   rj   z'Data2VecVisionSdpaSelfAttention.forward  sB  € ð ð 	Ý×ÒØ”>Ô*ð Dð Dð Dñô ð ð $Ô)¨#¨2¨#Ô.ˆØC˜ÐC bÐC¨$Ô*BÐCÐCˆØ—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆØ—H’H˜]Ñ+Ô+×0Ò0°Ñ>Ô>×HÒHÈÈAÑNÔNˆ	Ø—j’j Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆàˆ	ØÔ*ð 	Ø&‰MˆF�EØ! T¤[Ô%;Ñ;¸UÀdÄkÔF\Ñ=\Ð]ˆKØ×3Ò3ØÐ5ÀÔ@SÐTUÔ@Vð 4ñ ô ˆIð
 "Ð-ØÐ Ø2�	�	àÐ3Ñ3�	à•d”i Ô 8Ñ9Ô9Ñ9ˆÝœÔ+×HÒHØØØØØBFÄ-ÐX�d”kÔ>Ð>ÐUXØØð Iñ 
ô 
ˆð &×-Ò-¨a°°A°qÑ9Ô9×DÒDÑFÔFˆØ"/×"4Ò"4Ñ"6Ô"6°s¸°sÔ";¸tÔ?QÐ>SÑ"SÐØ*˜Ô*Ð,CÐDˆØ˜dÐ"Ð"r   r­   )	r   r   r   r(   rl   r’   r®   rm   rj   r   r   r   r°   r°     s¡   € € € € € ð #(Ø6:Ø).Ø(,ð/#ð /#à”|ð/#ð  ð/#ð !&¤¨tÑ 3ð	/#ð
 #'ð/#ð ˜#”J Ñ%ð/#ð 
ˆuŒ|Ô	˜u U¤\°5´<Ð%?Ô@Ñ	@ð/#ð /#ð /#ð /#ð /#ð /#r   r°   c                   ó`   ‡ — e Zd ZdZdeddfˆ fd„Zd	dej        dej        dej        fd„Zˆ xZ	S )
ÚData2VecVisionSelfOutputz«
    The residual connection is defined in Data2VecVisionLayer instead of here (as is the case with other models), due to the
    layernorm applied before each block.
    r"   r#   Nc                 óÌ   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          j        |j        ¦  «        | _        d S rk   )	r%   r&   r   r�   r*   Údenser9   r:   r;   ©r<   r"   r=   s     €r   r&   z!Data2VecVisionSelfOutput.__init__G  sJ   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜vÔ1°6Ô3EÑFÔFˆŒ
Ý”z &Ô"<Ñ=Ô=ˆŒˆˆr   r—   Úinput_tensorc                 óZ   — |                       |¦  «        }|                      |¦  «        }|S rk   ©rÀ   r;   )r<   r—   rÂ   Úgammas       r   rj   z Data2VecVisionSelfOutput.forwardL  ó*   € ØŸ
š
 =Ñ1Ô1ˆØŸš ]Ñ3Ô3ˆàÐr   rk   )
r   r   r   r   r   r&   r(   rl   rj   ro   rp   s   @r   r¾   r¾   A  sŒ   ø€ € € € € ðð ð
>Ð3ð >¸ð >ð >ð >ð >ð >ð >ð
ð  U¤\ð ÀÄð Ð^cÔ^jð ð ð ð ð ð ð ð r   r¾   )ÚeagerÚsdpac                   óÆ   ‡ — e Zd Zddededz  ddfˆ fd„Z	 	 	 	 ddej        ded	e	d
         dedee
         dz  deej                 eej        ej        f         z  fd„Zˆ xZS )ÚData2VecVisionAttentionNr"   rƒ   r#   c                 ó²   •— t          ¦   «                              ¦   «          t          |j                 ||¬¦  «        | _        t          |¦  «        | _        d S )Nrˆ   )r%   r&   Ú&DATA2VEC_VISION_SELF_ATTENTION_CLASSESÚ_attn_implementationÚ	attentionr¾   Úoutputr–   s      €r   r&   z Data2VecVisionAttention.__init__[  sQ   ø€ Ý‰Œ×ÒÑÔÐÝ?ÀÔ@[Ô\Ø ð
ñ 
ô 
ˆŒõ /¨vÑ6Ô6ˆŒˆˆr   Fr—   r˜   r•   r”   rZ   r™   c                 óŒ   — |                       |||||¦  «        }|                      |d         |¦  «        }|f|dd …         z   }|S )Nr   r   )rÎ   rÏ   )	r<   r—   r˜   r•   rZ   r™   Úself_outputsÚattention_outputr¬   s	            r   rj   zData2VecVisionAttention.forwardb  s[   € ð —~’~ØÐ,Ð.DÐF^Ð`jñ
ô 
ˆð  Ÿ;š; |°A¤¸ÑFÔFÐà#Ð%¨°Q°R°RÔ(8Ñ8ˆØˆr   rk   r­   )r   r   r   r   r®   r&   r(   rl   r’   r   rm   rj   ro   rp   s   @r   rÊ   rÊ   Z  sé   ø€ € € € € ð7ð 7Ð3ð 7À%È$Á,ð 7ÐZ^ð 7ð 7ð 7ð 7ð 7ð 7ð #(ØQUØ).Ø(,ðð à”|ðð  ðð !)Ð)MÔ Nð	ð
 #'ðð ˜#”J Ñ%ðð 
ˆuŒ|Ô	˜u U¤\°5´<Ð%?Ô@Ñ	@ðð ð ð ð ð ð ð r   rÊ   c                   óL   ‡ — e Zd Zdeddfˆ fd„Zdej        dej        fd„Zˆ xZS )ÚData2VecVisionIntermediater"   r#   Nc                 ó  •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          |j        t          ¦  «        rt          |j                 | _        d S |j        | _        d S rk   )r%   r&   r   r�   r*   Úintermediate_sizerÀ   r1   Ú
hidden_actÚstrr	   Úintermediate_act_fnrÁ   s     €r   r&   z#Data2VecVisionIntermediate.__init__v  sn   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜vÔ1°6Ô3KÑLÔLˆŒ
Ý�fÔ'­Ñ-Ô-ð 	9Ý'-¨fÔ.?Ô'@ˆDÔ$Ð$Ð$à'-Ô'8ˆDÔ$Ð$Ð$r   r—   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S rk   )rÀ   rÙ   ©r<   r—   s     r   rj   z"Data2VecVisionIntermediate.forward~  s,   € ØŸ
š
 =Ñ1Ô1ˆØ×0Ò0°Ñ?Ô?ˆàÐr   ©	r   r   r   r   r&   r(   rl   rj   ro   rp   s   @r   rÔ   rÔ   u  sr   ø€ € € € € ð9Ð3ð 9¸ð 9ð 9ð 9ð 9ð 9ð 9ð U¤\ð °e´lð ð ð ð ð ð ð ð r   rÔ   c                   óL   ‡ — e Zd Zdeddfˆ fd„Zdej        dej        fd„Zˆ xZS )ÚData2VecVisionOutputr"   r#   Nc                 óÌ   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          j        |j        ¦  «        | _	        d S rk   )
r%   r&   r   r�   rÖ   r*   rÀ   r9   r:   r;   rÁ   s     €r   r&   zData2VecVisionOutput.__init__‡  sJ   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜vÔ7¸Ô9KÑLÔLˆŒ
Ý”z &Ô"<Ñ=Ô=ˆŒˆˆr   r—   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S rk   rÄ   rÛ   s     r   rj   zData2VecVisionOutput.forwardŒ  rÆ   r   rÜ   rp   s   @r   rÞ   rÞ   †  sr   ø€ € € € € ð>Ð3ð >¸ð >ð >ð >ð >ð >ð >ð
 U¤\ð °e´lð ð ð ð ð ð ð ð r   rÞ   c                   ó^   ‡ — e Zd ZdZd
deddfˆ fd„Zdej        dej        fd„Zde	fd	„Z
ˆ xZS )ÚData2VecVisionDropPathzÏStochastic depth (DropPath) per sample, for residual blocks.

    Identity when ``drop_prob`` is 0 or outside training. See `Deep Networks with Stochastic Depth
    <https://arxiv.org/abs/1603.09382>`_.
    r²   Ú	drop_probr#   Nc                 óV   •— t          ¦   «                              ¦   «          || _        d S rk   )r%   r&   rã   )r<   rã   r=   s     €r   r&   zData2VecVisionDropPath.__init__›  s$   ø€ Ý‰Œ×ÒÑÔÐØ"ˆŒˆˆr   r—   c                 ó  — | j         dk    s| j        s|S d| j         z
  }|j        d         fd|j        dz
  z  z   }t	          j        ||j        |j        ¬¦  «        }t	          j        ||z   ¦  «        }| 	                    |¦  «        |z  S )Nr²   r   r   ©r   )r~   Údevice)
rã   rº   rK   Úndimr(   Úrandr~   rç   ÚfloorÚdiv)r<   r—   Ú	keep_probrK   Úrandom_tensors        r   rj   zData2VecVisionDropPath.forwardŸ  s“   € ØŒ>˜SÒ Ð ¨¬Ð Ø Ð Ø˜œÑ&ˆ	ØÔ$ QÔ'Ð)¨D°MÔ4FÈÑ4JÑ,KÑKˆÝœ
 5°Ô0CÈMÔL`ÐaÑaÔaˆÝœ M°IÑ$=Ñ>Ô>ˆØ× Ò  Ñ+Ô+¨mÑ;Ð;r   c                 ó   — d| j         › �S )Nzp=)rã   ©r<   s    r   Ú
extra_reprz!Data2VecVisionDropPath.extra_repr¨  s   € Ø$�D”NÐ$Ð$Ð$r   )r²   )r   r   r   r   Úfloatr&   r(   rl   rj   rØ   rð   ro   rp   s   @r   râ   râ   ”  s›   ø€ € € € € ðð ð#ð # %ð #°$ð #ð #ð #ð #ð #ð #ð< U¤\ð <°e´lð <ð <ð <ð <ð%˜Cð %ð %ð %ð %ð %ð %ð %ð %r   râ   c                   óØ   ‡ — e Zd ZdZ	 ddededz  deddfˆ fd„Z	 	 	 	 dd
ej	        de
dej	        dz  de
deeef         dz  deej	                 eej	        ej	        f         z  fd„Zˆ xZS )ÚData2VecVisionLayerz?This corresponds to the Block class in the timm implementation.Nr²   r"   rƒ   Údrop_path_rater#   c                 óè  •— t          ¦   «                              ¦   «          |j        | _        d| _        t	          ||¬¦  «        | _        t          |¦  «        | _        t          |¦  «        | _	        t          j        |j        |j        ¬¦  «        | _        |dk    rt          |¦  «        nt          j        ¦   «         | _        t          j        |j        |j        ¬¦  «        | _        |j        }|dk    rlt          j        |t+          j        |j        ¦  «        z  d¬¦  «        | _        t          j        |t+          j        |j        ¦  «        z  d¬¦  «        | _        d S d\  | _        | _        d S )	Nr   rˆ   ©Úepsr²   r   T)Úrequires_grad)NN)r%   r&   Úchunk_size_feed_forwardÚseq_len_dimrÊ   rÎ   rÔ   ÚintermediaterÞ   rÏ   r   Ú	LayerNormr*   Úlayer_norm_epsÚlayernorm_beforerâ   ÚIdentityÚ	drop_pathÚlayernorm_afterÚlayer_scale_init_valuer'   r(   ÚonesÚlambda_1Úlambda_2)r<   r"   rƒ   rô   Úinit_valuesr=   s        €r   r&   zData2VecVisionLayer.__init__°  s>  ø€ õ 	‰Œ×ÒÑÔÐØ'-Ô'EˆÔ$ØˆÔÝ0°À[ÐQÑQÔQˆŒÝ6°vÑ>Ô>ˆÔÝ*¨6Ñ2Ô2ˆŒÝ "¤¨VÔ-?ÀVÔEZÐ [Ñ [Ô [ˆÔØCQÐTWÒCWÐCWÕ/°Ñ?Ô?Ð?Õ]_Ô]hÑ]jÔ]jˆŒÝ!œ|¨FÔ,>ÀFÔDYÐZÑZÔZˆÔàÔ3ˆØ˜Š?ˆ?ÝœL¨µu´zÀ&ÔBTÑ7UÔ7UÑ)UÐeiÐjÑjÔjˆDŒMÝœL¨µu´zÀ&ÔBTÑ7UÔ7UÑ)UÐeiÐjÑjÔjˆDŒMˆMˆMà+5Ñ(ˆDŒM˜4œ=˜=˜=r   Fr—   r˜   r•   rZ   r™   c                 ó²  — |                       |                      |¦  «        ||||¬¦  «        }|d         }|dd …         }| j        �
| j        |z  }|                      |¦  «        |z   }|                      |¦  «        }	|                      |	¦  «        }	|                      |	¦  «        }	| j        �
| j        |	z  }	|                      |	¦  «        |z   }	|	f|z   }|S )N©r˜   r•   rZ   r™   r   r   )rÎ   rþ   r  r   r  rû   rÏ   r  )
r<   r—   r˜   r•   rZ   r™   Úself_attention_outputsrÒ   r¬   Úlayer_outputs
             r   rj   zData2VecVisionLayer.forwardÄ  sü   € ð "&§¢Ø×!Ò! -Ñ0Ô0Ø/Ø#9Ø%=Ø!ð "0ñ "
ô "
Ðð 2°!Ô4ÐØ(¨¨¨Ô,ˆð Œ=Ð$Ø#œ}Ð/?Ñ?Ðð ŸšÐ'7Ñ8Ô8¸=ÑHˆð ×+Ò+¨MÑ:Ô:ˆà×(Ò(¨Ñ6Ô6ˆØ—{’{ <Ñ0Ô0ˆàŒ=Ð$Øœ=¨<Ñ7ˆLð —~’~ lÑ3Ô3°mÑCˆà�/ GÑ+ˆàˆr   )Nr²   r­   )r   r   r   r   r   r®   rñ   r&   r(   rl   r’   rm   rj   ro   rp   s   @r   ró   ró   ­  s  ø€ € € € € ØIÐIð gjð6ð 6Ø*ð6Ø9>À¹ð6Ø^cð6à	ð6ð 6ð 6ð 6ð 6ð 6ð. #(Ø6:Ø).Ø-1ð'ð 'à”|ð'ð  ð'ð !&¤¨tÑ 3ð	'ð
 #'ð'ð ˜#˜s˜(”O dÑ*ð'ð 
ˆuŒ|Ô	˜u U¤\°5´<Ð%?Ô@Ñ	@ð'ð 'ð 'ð 'ð 'ð 'ð 'ð 'r   ró   c                   ó¦   ‡ — e Zd Zdededdfˆ fd„Ze ed¬¦  «        deeef         de	j
        fd„¦   «         ¦   «         Zdd
ede	j
        fd„Zˆ xZS )r”   r"   rƒ   r#   Nc                 ó  •— t          ¦   «                              ¦   «          || _        d|d         z  dz
  d|d         z  dz
  z  dz   | _        t	          j        t          j        | j        |j        ¦  «        ¦  «        | _	        d S )NrC   r   r   r   )
r%   r&   rƒ   Únum_relative_distancer   r'   r(   r)   r‰   Úrelative_position_bias_tabler–   s      €r   r&   z+Data2VecVisionRelativePositionBias.__init__ð  s�   ø€ Ý‰Œ×ÒÑÔÐØ&ˆÔØ&'¨+°a¬.Ñ&8¸1Ñ&<ÀÀ[ÐQRÄ^ÑASÐVWÑAWÑ%XÐ[\Ñ%\ˆÔ"Ý,.¬LÝŒK˜Ô2°FÔ4NÑOÔOñ-
ô -
ˆÔ)Ð)Ð)r   é
   )Úmaxsizec                 ó^  — d| d         z  dz
  d| d         z  dz
  z  dz   }| d         | d         z  }t          j        t          j        | d         ¦  «        t          j        | d         ¦  «        d¬¦  «        }t          j        |¦  «        }t          j        |d¦  «        }|dd…dd…df         |dd…ddd…f         z
  }|                     ddd¦  «                             ¦   «         }|dd…dd…dfxx         | d         dz
  z  cc<   |dd…dd…dfxx         | d         dz
  z  cc<   |dd…dd…dfxx         d| d         z  dz
  z  cc<   t          j        |dz   fdz  |j        ¬¦  «        }| 	                    d	¦  «        |dd…dd…f<   |dz
  |ddd…f<   |dz
  |dd…df<   |dz
  |d
<   |S )z¿
        This method creates the relative position index, modified to support arbitrary window sizes,
        as introduced in [MiDaS v3.1](https://huggingface.co/papers/2307.14460).
        rC   r   r   r   Úij)ÚindexingN)rF   r~   rB   )r   r   )
r(   ÚmeshgridÚarangeÚstackr   rO   r¢   r)   r~   Úsum)rƒ   r  Úwindow_areaÚgridÚcoordsÚcoords_flattenÚrelative_coordsÚrelative_position_indexs           r   Ú generate_relative_position_indexzCData2VecVisionRelativePositionBias.generate_relative_position_indexù  s  € ð "# [°¤^Ñ!3°aÑ!7¸AÀÈAÄÑ<NÐQRÑ<RÑ SÐVWÑ WÐð " !”n {°1¤~Ñ5ˆÝŒ~�eœl¨;°q¬>Ñ:Ô:½E¼LÈÐUVÌÑ<XÔ<XÐcgÐhÑhÔhˆÝ”˜TÑ"Ô"ˆÝœ v¨qÑ1Ô1ˆØ(¨¨¨¨A¨A¨A¨t¨Ô4°~ÀaÀaÀaÈÈqÈqÈqÀjÔ7QÑQˆØ)×1Ò1°!°Q¸Ñ:Ô:×EÒEÑGÔGˆØ˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  K°¤N°QÑ$6Ñ6Ð Ð Ñ Ø˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  K°¤N°QÑ$6Ñ6Ð Ð Ñ Ø˜˜˜˜1˜1˜1˜a˜Ð Ð Ô  A¨°A¬Ñ$6¸Ñ$:Ñ:Ð Ð Ñ Ý"'¤+°KÀ!±OÐ3EÈÑ3IÐQ`ÔQfÐ"gÑ"gÔ"gÐØ*9×*=Ò*=¸bÑ*AÔ*AÐ    A B B Ñ'Ø)>ÀÑ)BÐ  1 2 2 Ñ&Ø)>ÀÑ)BÐ    A Ñ&Ø(=ÀÑ(AÐ Ñ%Ø&Ð&r   FrZ   c                 ó0  — d| j         d         z  dz
  }d| j         d         z  dz
  }d|d         z  dz
  }d|d         z  dz
  }| j        }| j        }	||z  dz   }
|d|	dz
  …         }|                     d||d¦  «                             dddd¦  «        }t
          j                             |t          |¦  «        t          |¦  «        fd¬¦  «        }|                     dddd¦  «                             |
dz
  d¦  «        }t          j
        |||	dz
  d…         g¦  «        }|                      |¦  «        }||                     d¦  «                 }|                     |d         |d         z  dz   |d         |d         z  dz   d¦  «        }|                     ddd¦  «                             ¦   «         }|rKt
          j                             |                     d¦  «        ||fdd	¬
¦  «                             d¦  «        }|                     d¦  «        S )zu
        Modification of timm.models.beit.py: Attention._get_rel_pos_bias to support arbitrary window sizes.
        rC   r   r   r   NrB   Úbilinear)rF   rG   FrE   )rƒ   r  r  rN   rO   r   rP   rQ   r   r(   rS   r  rR   r¢   r`   Úsqueeze)r<   rƒ   rZ   r�   Ú
old_heightÚ	old_widthrW   rX   Ú old_relative_position_bias_tableÚold_num_relative_distanceÚnew_num_relative_distanceÚold_sub_tableÚnew_sub_tableÚ new_relative_position_bias_tabler  r•   s                   r   rj   z*Data2VecVisionRelativePositionBias.forward  s`  € ð ˜Ô)¨!Ô,Ñ,¨qÑ0ˆ
Ø˜Ô(¨Ô+Ñ+¨aÑ/ˆ	à˜ QœÑ'¨!Ñ+ˆ
Ø˜ AœÑ&¨Ñ*ˆ	à+/Ô+LÐ(à$(Ô$>Ð!Ø$.°Ñ$:¸QÑ$>Ð!à8Ð9XÐ;TÐWXÑ;XÐ9XÔYˆà%×-Ò-¨a°¸JÈÑKÔK×SÒSÐTUÐWXÐZ[Ð]^Ñ_Ô_ˆÝœ×1Ò1Ø¥¨:Ñ!6Ô!6½	À)Ñ8LÔ8LÐ MÐT^ð 2ñ 
ô 
ˆð &×-Ò-¨a°°A°qÑ9Ô9×AÒAÐB[Ð^_ÑB_ÐacÑdÔdˆå+0¬9ØÐ<Ð=VÐYZÑ=ZÐ=\Ð=\Ô]Ð^ñ,
ô ,
Ð(ð #'×"GÒ"GÈÑ"TÔ"TÐØ!AÐBY×B^ÒB^Ð_aÑBbÔBbÔ!cÐð "8×!<Ò!<Ø˜ŒN˜[¨œ^Ñ+¨aÑ/°¸Q´À+ÈaÄ.Ñ1PÐSTÑ1TÐVXñ"
ô "
Ðð "8×!?Ò!?ÀÀ1ÀaÑ!HÔ!H×!SÒ!SÑ!UÔ!UÐà#ð 	Ý%'¤]×%>Ò%>Ø&×0Ò0°Ñ3Ô3Ø Ð)ØØ#ð	 &?ñ &ô &÷
 Šg�a‰jŒjð #ð &×/Ò/°Ñ2Ô2Ð2r   )FN)r   r   r   r   r®   r&   Ústaticmethodr   rm   r(   rl   r  r’   rj   ro   rp   s   @r   r”   r”   ï  sÏ   ø€ € € € € ð
Ð3ð 
À%ð 
ÈDð 
ð 
ð 
ð 
ð 
ð 
ð Ø(Ð(°Ð4Ñ4Ô4ð'°e¸CÀ¸H´oð 'È%Ì,ð 'ð 'ð 'ñ 5Ô4ñ „\ð'ð0-3ð -3¸Tð -3Ð]bÔ]ið -3ð -3ð -3ð -3ð -3ð -3ð -3ð -3r   r”   c                   óŠ   ‡ — e Zd Zddededz  ddfˆ fd„Z	 	 	 	 	 ddej        d	ed
ededee	e	f         dz  dedee
z  fd„Zˆ xZS )ÚData2VecVisionEncoderNr"   rƒ   r#   c                 ó†  •‡‡‡— t          ¦   «                              ¦   «          ‰| _        ‰j        | _        | j        rt          ‰‰¬¦  «        | _        d„ t          j        d‰j	        ‰j
        d¬¦  «        D ¦   «         Št          j        ˆˆˆfd„t          ‰j
        ¦  «        D ¦   «         ¦  «        | _        d| _        d S )Nrˆ   c                 ó6   — g | ]}|                      ¦   «         ‘ŒS r   )Úitem)Ú.0Úxs     r   ú
<listcomp>z2Data2VecVisionEncoder.__init__.<locals>.<listcomp>M  s    € ÐrÐrÐr˜Aˆq�vŠv‰xŒxÐrÐrÐrr   r   Úcpu)rç   c                 óR   •— g | ]#}t          ‰‰j        r‰nd ‰|         ¬¦  «        ‘Œ$S )N)rƒ   rô   )ró   Úuse_relative_position_bias)r0  Úir"   Údprrƒ   s     €€€r   r2  z2Data2VecVisionEncoder.__init__.<locals>.<listcomp>O  sT   ø€ ð ð ð ð õ $ØØ/5Ô/PÐ Z  ÐVZØ#& q¤6ðñ ô ðð ð r   F)r%   r&   r"   Ú!use_shared_relative_position_biasr“   r”   r•   r(   Úlinspacerô   Únum_hidden_layersr   Ú
ModuleListÚrangeÚlayerÚgradient_checkpointing)r<   r"   rƒ   r7  r=   s    ``@€r   r&   zData2VecVisionEncoder.__init__E  sà   øøøø€ Ý‰Œ×ÒÑÔÐØˆŒØ*0Ô*RˆÔ'ØÔ*ð 	nÝ*LÈVÐalÐ*mÑ*mÔ*mˆDÔ'ð sÐr¥¤°°6Ô3HÈ&ÔJbÐkpÐ!qÑ!qÔ!qÐrÑrÔrˆÝ”]ðð ð ð ð ð õ ˜vÔ7Ñ8Ô8ðñ ô ñ	
ô 	
ˆŒ
ð ',ˆÔ#Ð#Ð#r   FTr—   r˜   Úoutput_hidden_statesrZ   r™   Úreturn_dictc                 óº  — |rdnd }|rdnd }t          | j        ¦  «        D ]ƒ\  }	}
|r||fz   }| j        rG|\  }}|| j        j        z  || j        j        z  f}|                      |||j        d         ¬¦  «        }nd } |
|||||¬¦  «        }|d         }|r||d         fz   }Œ„|r||fz   }|st          d„ |||fD ¦   «         ¦  «        S t          |||¬¦  «        S )Nr   r   )rZ   r�   r  r   c              3   ó   K  — | ]}|®|V — Œ	d S rk   r   )r0  Úvs     r   ú	<genexpr>z0Data2VecVisionEncoder.forward.<locals>.<genexpr>„  s(   è è € ÐmÐm˜qÐ_`Ð_l˜Ð_lÐ_lÐ_lÐ_lÐmÐmr   )Úlast_hidden_stater—   Ú
attentions)	Ú	enumerater=  r“   r"   r0   r•   rK   r®   r   )r<   r—   r˜   r?  rZ   r™   r@  Úall_hidden_statesÚall_self_attentionsr6  Úlayer_moduler?   r@   rƒ   r•   Úlayer_outputss                   r   rj   zData2VecVisionEncoder.forwardZ  sw  € ð #7Ð@˜B˜B¸DÐØ$5Ð?˜b˜b¸4Ðå(¨¬Ñ4Ô4ð 	Pð 	P‰OˆAˆ|Ø#ð IØ$5¸Ð8HÑ$HÐ!àÔ.ð .Ø *‘�˜Ø%¨¬Ô)?Ñ?ÀÈ$Ì+ÔJ`ÑA`Ða�Ø)-×)DÒ)DØÐ:RÐ]jÔ]pÐqrÔ]sð *Eñ *ô *Ð&Ð&ð *.Ð&à(˜LØØ"3Ø'=Ø)AØ%ðñ ô ˆMð *¨!Ô,ˆMà ð PØ&9¸]È1Ô=MÐ<OÑ&OÐ#øàð 	EØ 1°]Ð4DÑ DÐàð 	nÝÐmÐm ]Ð4EÐGZÐ$[ÐmÑmÔmÑmÔmÐmÝØ+Ø+Ø*ð
ñ 
ô 
ð 	
r   rk   )FFFNT)r   r   r   r   r®   r&   r(   rl   r’   rm   r   rj   ro   rp   s   @r   r,  r,  D  sß   ø€ € € € € ð,ð ,Ð3ð ,À%È$Á,ð ,ÐZ^ð ,ð ,ð ,ð ,ð ,ð ,ð0 #(Ø%*Ø).Ø-1Ø ð/
ð /
à”|ð/
ð  ð/
ð #ð	/
ð
 #'ð/
ð ˜#˜s˜(”O dÑ*ð/
ð ð/
ð 
�Ñ	 ð/
ð /
ð /
ð /
ð /
ð /
ð /
ð /
r   r,  c                   ót   ‡ — e Zd ZU eed<   dZdZdZdZdgZ	dgZ
dZ ej        ¦   «         ˆ fd„¦   «         Zˆ xZS )	ÚData2VecVisionPreTrainedModelr"   Údata2vec_vision)Úimager[   Tró   z.*relative_position_index.*c                 ó^  •— t          ¦   «                              |¦  «         t          |t          ¦  «        r]t	          j        |j        ¦  «         |j        �t	          j        |j        ¦  «         |j        �t	          j        |j        ¦  «         dS dS t          |t          ¦  «        rt	          j        |j
        ¦  «         dS t          |t          ¦  «        rQ|j        �Lt	          j        |j        | j        j        ¦  «         t	          j        |j        | j        j        ¦  «         dS dS dS )zInitialize the weightsN)r%   Ú_init_weightsr1   r!   ÚinitÚzeros_r+   r-   r8   r”   r  ró   r  Ú	constant_r"   r  r  )r<   Úmoduler=   s     €r   rQ  z+Data2VecVisionPreTrainedModel._init_weights˜  s  ø€ õ 	‰Œ×Ò˜fÑ%Ô%Ð%Ý�fÕ6Ñ7Ô7ð 	TÝŒK˜Ô(Ñ)Ô)Ð)ØÔ Ð,Ý”˜FÔ-Ñ.Ô.Ð.ØÔ)Ð5Ý”˜FÔ6Ñ7Ô7Ð7Ð7Ð7ð 6Ð5å˜Õ BÑCÔCð 	TÝŒK˜Ô;Ñ<Ô<Ð<Ð<Ð<Ý˜Õ 3Ñ4Ô4ð 	TØŒÐ*Ý”˜vœ°´Ô0RÑSÔSÐSÝ”˜vœ°´Ô0RÑSÔSÐSÐSÐSð	Tð 	TØ*Ð*r   )r   r   r   r   Ú__annotations__Úbase_model_prefixÚinput_modalitiesÚmain_input_nameÚsupports_gradient_checkpointingÚ_no_split_modulesÚ"_keys_to_ignore_on_load_unexpectedÚ_supports_sdpar(   Úno_gradrQ  ro   rp   s   @r   rM  rM  Œ  s‘   ø€ € € € € € ð !Ð Ð Ñ Ø)ÐØ!ÐØ$€OØ&*Ð#Ø.Ð/ÐØ*HÐ)IÐ&Ø€Nà€U„]�_„_ðTð Tð Tð Tñ „_ðTð Tð Tð Tð Tr   rM  c                   ó¦   ‡ — e Zd Zddededdfˆ fd„Zd„ Ze	 	 	 	 	 ddej	        d	ej
        dz  d
edz  dedz  dededz  deez  fd„¦   «         Zˆ xZS )ÚData2VecVisionModelFr"   Úadd_pooling_layerr#   Nc                 ó   •— t          ¦   «                              |¦  «         || _        t          |¦  «        | _        t          || j        j        j        ¬¦  «        | _        |j	        rt          j        ¦   «         nt          j        |j        |j        ¬¦  «        | _        |rt!          |¦  «        nd| _        |                      ¦   «          dS )zw
        add_pooling_layer (bool, *optional*, defaults to `False`):
            Whether to add a pooling layer
        rˆ   rö   N)r%   r&   r"   r!   r>   r,  r/   rw   ÚencoderÚuse_mean_poolingr   rÿ   rü   r*   rý   Ú	layernormÚData2VecVisionPoolerÚpoolerÚ	post_init)r<   r"   ra  r=   s      €r   r&   zData2VecVisionModel.__init__­  s¸   ø€ õ
 	‰Œ×Ò˜Ñ Ô Ð ØˆŒå2°6Ñ:Ô:ˆŒÝ,¨VÀÄÔAaÔAmÐnÑnÔnˆŒð $Ô4Ðu�BŒK‰MŒMˆM½"¼,ÀvÔGYÐ_eÔ_tÐ:uÑ:uÔ:uð 	Œð 7HÐQÕ*¨6Ñ2Ô2Ð2ÈTˆŒð 	�ŠÑÔÐÐÐr   c                 ó   — | j         j        S rk   )r>   r/   rï   s    r   Úget_input_embeddingsz(Data2VecVisionModel.get_input_embeddingsÀ  s   € ØŒÔ/Ð/r   r[   r\   r˜   r?  rZ   r@  c                 óÊ  — |�|n| j         j        }|�|n| j         j        }|�|n| j         j        }|                      ||¬¦  «        \  }}	|j        dd…         }
|                      ||||
||¬¦  «        }|d         }|                      |¦  «        }| j        �|                      |¦  «        nd}|s|�||fn|f}||dd…         z   S t          |||j
        |j        ¬¦  «        S )zË
        bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, num_patches)`, *optional*):
            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
        N)r\   rC   )r˜   r?  r™   r@  rZ   r   r   )rE  Úpooler_outputr—   rF  )r"   r˜   r?  r@  r>   rK   rc  re  rg  r   r—   rF  )r<   r[   r\   r˜   r?  rZ   r@  ÚkwargsÚembedding_outputrb   r™   Úencoder_outputsÚsequence_outputÚpooled_outputÚhead_outputss                  r   rj   zData2VecVisionModel.forwardÃ  s8  € ð 2CÐ1NÐ-Ð-ÐTXÔT_ÔTqÐà$8Ð$DÐ Ð È$Ì+ÔJjð 	ð &1Ð%<�k�kÀ$Ä+ÔBYˆà"Ÿošo¨lÈO˜oÑ\Ô\ÑÐ˜!Ø!Ô'¨¨¨Ô+ˆ
àŸ,š,ØØ/Ø!5Ø!Ø#Ø%=ð 'ñ 
ô 
ˆð *¨!Ô,ˆØŸ.š.¨Ñ9Ô9ˆØ8<¼Ð8O˜Ÿš OÑ4Ô4Ð4ÐUYˆàð 	6Ø?LÐ?X˜O¨]Ð;Ð;Ð_nÐ^pˆLØ /°!°"°"Ô"5Ñ5Ð5å3Ø-Ø'Ø)Ô7Ø&Ô1ð	
ñ 
ô 
ð 	
r   )F)NNNFN)r   r   r   r   r’   r&   rj  r   r(   rl   rn   r®   r   rj   ro   rp   s   @r   r`  r`  ª  sþ   ø€ € € € € ðð Ð3ð Èð ÐY]ð ð ð ð ð ð ð&0ð 0ð 0ð ð 48Ø)-Ø,0Ø).Ø#'ð,
ð ,
à”lð,
ð Ô)¨DÑ0ð,
ð   $™;ð	,
ð
 # T™kð,
ð #'ð,
ð ˜D‘[ð,
ð 
Ð5Ñ	5ð,
ð ,
ð ,
ñ „^ð,
ð ,
ð ,
ð ,
ð ,
r   r`  c                   óL   ‡ — e Zd Zdeddfˆ fd„Zdej        dej        fd„Zˆ xZS )rf  r"   r#   Nc                 ó¤   •— t          ¦   «                              ¦   «          |j        r t          j        |j        |j        ¬¦  «        nd | _        d S )Nrö   )r%   r&   rd  r   rü   r*   rý   re  rÁ   s     €r   r&   zData2VecVisionPooler.__init__õ  sJ   ø€ Ý‰Œ×ÒÑÔÐàKQÔKbÐl�BŒL˜Ô+°Ô1FÐGÑGÔGÐGÐhlð 	Œˆˆr   r—   c                 ó–   — | j         �7|                       |d d …dd …d d …f                              d¦  «        ¦  «        n|d d …df         S )Nr   r   )re  ÚmeanrÛ   s     r   rj   zData2VecVisionPooler.forwardû  sX   € àBFÄ.ÐB\ˆt�~Š~˜m¨A¨A¨A¨q¨r¨r°1°1°1¨HÔ5×:Ò:¸1Ñ=Ô=Ñ>Ô>Ð>ÐboÐpqÐpqÐpqÐstÐptÔbuÐur   rÜ   rp   s   @r   rf  rf  ô  s|   ø€ € € € € ð
Ð3ð 
¸ð 
ð 
ð 
ð 
ð 
ð 
ðv U¤\ð v°e´lð vð vð vð vð vð vð vð vr   rf  zÀ
    Data2VecVision Model transformer with an image classification head on top (a linear layer on top of the average of
    the final hidden states of the patch tokens) e.g. for ImageNet.
    c                   ó¢   ‡ — e Zd Zdeddfˆ fd„Ze	 	 	 	 	 	 ddej        dz  dej        dz  dedz  d	edz  d
ededz  de	e
z  fd„¦   «         Zˆ xZS )Ú$Data2VecVisionForImageClassificationr"   r#   Nc                 ó:  •— t          ¦   «                              |¦  «         |j        | _        t          |d¬¦  «        | _        |j        dk    rt          j        |j        |j        ¦  «        nt          j        ¦   «         | _	        |  
                    ¦   «          d S )NT©ra  r   )r%   r&   Ú
num_labelsr`  rN  r   r�   r*   rÿ   Ú
classifierrh  rÁ   s     €r   r&   z-Data2VecVisionForImageClassification.__init__  sŒ   ø€ Ý‰Œ×Ò˜Ñ Ô Ð à Ô+ˆŒÝ2°6ÈTÐRÑRÔRˆÔð OUÔN_ÐbcÒNcÐNc�"œ) FÔ$6¸Ô8IÑJÔJÐJÕikÔitÑivÔivˆŒð 	�ŠÑÔÐÐÐr   Fr[   Úlabelsr˜   r?  rZ   r@  c                 óP  — |�|n| j         j        }|                      |||||¬¦  «        }|r|j        n|d         }	|                      |	¦  «        }
d}|�|                      ||
| j         ¦  «        }|s|
f|dd…         z   }|�|f|z   n|S t          ||
|j        |j        ¬¦  «        S )aŠ  
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        N©r˜   r?  rZ   r@  r   rC   ©ÚlossÚlogitsr—   rF  )	r"   r@  rN  rl  r|  Úloss_functionr   r—   rF  )r<   r[   r}  r˜   r?  rZ   r@  rm  r¬   rq  r‚  r�  rÏ   s                r   rj   z,Data2VecVisionForImageClassification.forward  sê   € ð" &1Ð%<�k�kÀ$Ä+ÔBYˆØ×&Ò&ØØ/Ø!5Ø%=Ø#ð 'ñ 
ô 
ˆð 2=ÐL˜Ô-Ð-À'È!Ä*ˆà—’ Ñ/Ô/ˆàˆØÐØ×%Ò% f¨f°d´kÑBÔBˆDàð 	FØ�Y ¨¨¨¤Ñ,ˆFØ)-Ð)9�T�G˜fÑ$Ð$¸vÐEå$ØØØ!Ô/ØÔ)ð	
ñ 
ô 
ð 	
r   ©NNNNFN)r   r   r   r   r&   r   r(   rl   r’   r®   r   rj   ro   rp   s   @r   rx  rx     sé   ø€ € € € € ð
Ð3ð 
¸ð 
ð 
ð 
ð 
ð 
ð 
ð ð -1Ø&*Ø)-Ø,0Ø).Ø#'ð*
ð *
à”l TÑ)ð*
ð ”˜tÑ#ð*
ð   $™;ð	*
ð
 # T™kð*
ð #'ð*
ð ˜D‘[ð*
ð 
Ð&Ñ	&ð*
ð *
ð *
ñ „^ð*
ð *
ð *
ð *
ð *
r   rx  c                   óÀ   ‡ — e Zd Z	 	 	 	 	 	 	 ddededeeeef         z  d	ed
eeeef         z  ez  dedeeeef         z  dedefˆ fd„Zdej	        dej	        fd„Z
ˆ xZS )ÚData2VecVisionConvLayerr   r   r   FÚreluÚin_channelsÚout_channelsrt   ru   Úpaddingr‡   ÚdilationÚgroupsÚ
activationc
           
      ó
  •— t          ¦   «                              ¦   «          t          j        ||||||||¬¦  «        | _        t          j        |¦  «        | _        |	�t          |	         nt          j        ¦   «         | _	        d S )N)rˆ  r‰  rt   ru   rŠ  r‹  rŒ  r‡   )
r%   r&   r   rx   ÚconvolutionÚBatchNorm2dÚnormalizationr	   rÿ   r�  )r<   rˆ  r‰  rt   ru   rŠ  r‡   r‹  rŒ  r�  r=   s             €r   r&   z Data2VecVisionConvLayer.__init__D  s€   ø€ õ 	‰Œ×ÒÑÔÐÝœ9Ø#Ø%Ø#ØØØØØð	
ñ 	
ô 	
ˆÔõ  œ^¨LÑ9Ô9ˆÔØ0:Ð0F�& Ô,Ð,ÍBÌKÉMÌMˆŒˆˆr   r—   r#   c                 ó„   — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|S rk   )r�  r‘  r�  rÛ   s     r   rj   zData2VecVisionConvLayer.forward^  s?   € Ø×(Ò(¨Ñ7Ô7ˆØ×*Ò*¨=Ñ9Ô9ˆØŸš¨Ñ6Ô6ˆØÐr   )r   r   r   Fr   r   r‡  )r   r   r   rm   r®   rØ   r’   r&   r(   rl   rj   ro   rp   s   @r   r†  r†  C  s  ø€ € € € € ð
 ./ØØ/0ØØ*+ØØ ðZð ZàðZð ðZð ˜5  c œ?Ñ*ð	Zð
 ðZð �u˜S #˜X”Ñ&¨Ñ,ðZð ðZð ˜˜c 3˜hœÑ'ðZð ðZð ðZð Zð Zð Zð Zð Zð4 U¤\ð °e´lð ð ð ð ð ð ð ð r   r†  c                   óh   ‡ — e Zd Zdedededdfˆ fd„Zdej        deeef         dej        fd	„Zˆ xZ	S )
Ú!Data2VecVisionPyramidPoolingBlockÚ
pool_scalerˆ  Úchannelsr#   Nc                 ó¨   •— t          ¦   «                              ¦   «          t          j        |¦  «        | _        t          ||d¬¦  «        | _        d S )Nr   ©rt   )r%   r&   r   ÚAdaptiveAvgPool2dÚpoolingr†  Úconv)r<   r•  rˆ  r–  r=   s       €r   r&   z*Data2VecVisionPyramidPoolingBlock.__init__g  sE   ø€ Ý‰Œ×ÒÑÔÐÝÔ+¨JÑ7Ô7ˆŒÝ+¨K¸ÈqÐQÑQÔQˆŒ	ˆ	ˆ	r   ÚinputrF   c                 ó    — |                       |¦  «        }|                      |¦  «        }t          j                             ||dd¬¦  «        }|S )Nr   FrE   )rš  r›  r   rP   rQ   )r<   rœ  rF   Úhidden_states       r   rj   z)Data2VecVisionPyramidPoolingBlock.forwardl  sI   € Ø—|’| EÑ*Ô*ˆØ—y’y Ñ.Ô.ˆÝ”}×0Ò0°ÀDÈzÐinÐ0ÑoÔoˆØÐr   )
r   r   r   rm   r&   r(   rl   r®   rj   ro   rp   s   @r   r”  r”  f  s˜   ø€ € € € € ðR 3ð R°Sð RÀCð RÈDð Rð Rð Rð Rð Rð Rð
˜Uœ\ð °°s¸C°x´ð ÀUÄ\ð ð ð ð ð ð ð ð r   r”  c                   ót   ‡ — e Zd ZdZdeedf         dededdfˆ fd„Zd	ej        de	ej                 fd
„Z
ˆ xZS )Ú"Data2VecVisionPyramidPoolingModuleak  
    Pyramid Pooling Module (PPM) used in PSPNet.

    Args:
        pool_scales (tuple[int]): Pooling scales used in Pooling Pyramid
            Module.
        in_channels (int): Input channels.
        channels (int): Channels after modules, before conv_seg.

    Based on OpenMMLab's implementation, found in https://github.com/open-mmlab/mmsegmentation.
    Úpool_scales.rˆ  r–  r#   Nc                 óÂ   •‡‡— t          ¦   «                              ¦   «          || _        ‰| _        ‰| _        t          j        ˆˆfd„|D ¦   «         ¦  «        | _        d S )Nc                 ó4   •— g | ]}t          |‰‰¬ ¦  «        ‘ŒS ))r•  rˆ  r–  )r”  )r0  r•  r–  rˆ  s     €€r   r2  z?Data2VecVisionPyramidPoolingModule.__init__.<locals>.<listcomp>‡  s:   ø€ ð ð ð àõ 2¸ZÐU`ÐksÐtÑtÔtðð ð r   )r%   r&   r¡  rˆ  r–  r   r;  Úblocks)r<   r¡  rˆ  r–  r=   s     ``€r   r&   z+Data2VecVisionPyramidPoolingModule.__init__�  st   øøø€ Ý‰Œ×ÒÑÔÐØ&ˆÔØ&ˆÔØ ˆŒÝ”mðð ð ð ð à"-ðñ ô ñ
ô 
ˆŒˆˆr   r—   c                 óf   ‡‡— ‰                      ¦   «         dd …         Šˆˆfd„| j        D ¦   «         S )NrC   c                 ó*   •— g | ]} |‰‰¬ ¦  «        ‘ŒS ))rF   r   )r0  Úblockr—   Úoriginal_sizes     €€r   r2  z>Data2VecVisionPyramidPoolingModule.forward.<locals>.<listcomp>�  s(   ø€ ÐRÐRÐR¸U���m¨-Ð8Ñ8Ô8ÐRÐRÐRr   )rF   r¤  )r<   r—   r¨  s    `@r   rj   z*Data2VecVisionPyramidPoolingModule.forward�  s=   øø€ Ø%×*Ò*Ñ,Ô,¨Q¨R¨RÔ0ˆØRÐRÐRÐRÐRÀdÄkÐRÑRÔRÐRr   )r   r   r   r   r®   rm   r&   r(   rl   Úlistrj   ro   rp   s   @r   r   r   t  s¢   ø€ € € € € ð
ð 
ð

 E¨#¨s¨(¤Oð 

À#ð 

ÐQTð 

ÐY]ð 

ð 

ð 

ð 

ð 

ð 

ðS U¤\ð S°d¸5¼<Ô6Hð Sð Sð Sð Sð Sð Sð Sð Sr   r   c                   óŒ   ‡ — e Zd ZdZdeddfˆ fd„Zdeej                 dej        fd„Z	deej                 dej        fd	„Z
ˆ xZS )
ÚData2VecVisionUperHeadzú
    Unified Perceptual Parsing for Scene Understanding. This head is the implementation of
    [UPerNet](https://huggingface.co/papers/1807.10221).

    Based on OpenMMLab's implementation, found in https://github.com/open-mmlab/mmsegmentation.
    r"   r#   Nc           	      ó~  •— t          ¦   «                              ¦   «          |j        | _        |j        gdz  | _        |j        | _        t          j        | j        |j        d¬¦  «        | _	        t          | j        | j        d         | j        ¦  «        | _        t          | j        d         t          | j        ¦  «        | j        z  z   | j        dd¬¦  «        | _        t          j        ¦   «         | _        t          j        ¦   «         | _        | j        d d…         D ]f}| j                             t          || j        d¬¦  «        ¦  «         | j                             t          | j        | j        dd¬¦  «        ¦  «         Œgt          t          | j        ¦  «        | j        z  | j        dd¬¦  «        | _        d S )Né   r   r˜  rB   r   ©rt   rŠ  )r%   r&   r¡  r*   rˆ  r–  r   rx   r{  r|  r   Úpsp_modulesr†  ÚlenÚpsp_bottleneckr;  Úlateral_convsÚ	fpn_convsÚappendÚfpn_bottleneck)r<   r"   rˆ  r=   s      €r   r&   zData2VecVisionUperHead.__init__›  s¡  ø€ Ý‰Œ×ÒÑÔÐà!Ô-ˆÔØ"Ô.Ð/°!Ñ3ˆÔØÔ*ˆŒÝœ) D¤M°6Ô3DÐRSÐTÑTÔTˆŒõ >ØÔØÔ˜RÔ ØŒMñ
ô 
ˆÔõ
 6ØÔ˜RÔ ¥3 tÔ'7Ñ#8Ô#8¸4¼=Ñ#HÑHØŒMØØð	
ñ 
ô 
ˆÔõ  œ]™_œ_ˆÔÝœ™œˆŒØÔ+¨C¨R¨CÔ0ð 	sð 	sˆKØÔ×%Ò%Õ&=¸kÈ4Ì=ÐfgÐ&hÑ&hÔ&hÑiÔiÐiØŒN×!Ò!Õ"9¸$¼-ÈÌÐdeÐopÐ"qÑ"qÔ"qÑrÔrÐrÐrå5Ý�Ô Ñ!Ô! D¤MÑ1ØŒMØØð	
ñ 
ô 
ˆÔÐÐr   r—   c                 ó”   — |d         }t          j        |g|                      |¦  «        ¢d¬¦  «        }|                      |¦  «        S r^   )r(   rS   r¯  r±  )r<   r—   rž  s      r   Úpsp_forwardz"Data2VecVisionUperHead.psp_forward½  sK   € Ø$ RÔ(ˆÝ”y ,Ð!P°×1AÒ1AÀ,Ñ1OÔ1OÐ!PÐVWÐXÑXÔXˆØ×"Ò" <Ñ0Ô0Ð0r   Úencoder_hidden_statesc                 ó¬  — g }t          | j        |¦  «        D ]#\  }}|                      ||¦  «        ¦  «         Œ$|                     |                      |¦  «        ¦  «         t	          |¦  «        }t          |dz
  dd¦  «        D ]U}||dz
           j        dd …         }||dz
           t          j         	                    ||         |dd¬¦  «        z   ||dz
  <   ŒVg }t          |dz
  ¦  «        D ]1}|                      | j
        |         ||         ¦  «        ¦  «         Œ2|                     |d         ¦  «         t          |dz
  dd¦  «        D ]A}t          j         	                    ||         |d         j        dd …         dd¬¦  «        ||<   ŒBt          j        |d¬¦  «        }|                      |¦  «        }	|                      |	¦  «        }	|	S )	Nr   r   rB   rC   r   FrE   rI   )Úzipr²  r´  r·  r°  r<  rK   r   rP   rQ   r³  r(   rS   rµ  r|  )
r<   r¸  ÚlateralsÚlateral_convrž  Úused_backbone_levelsr6  Ú
prev_shapeÚfpn_outsrÏ   s
             r   rj   zData2VecVisionUperHead.forwardÂ  sù  € àˆÝ*-¨dÔ.@ÐBWÑ*XÔ*Xð 	8ð 	8Ñ&ˆL˜,Ø�OŠO˜L˜L¨Ñ6Ô6Ñ7Ô7Ð7Ð7à�Š˜×(Ò(Ð)>Ñ?Ô?Ñ@Ô@Ð@õ  # 8™}œ}ÐÝÐ+¨aÑ/°°BÑ7Ô7ð 	ð 	ˆAØ! ! a¡%œÔ.¨q¨r¨rÔ2ˆJØ& q¨1¡uœoµ´×0IÒ0IØ˜” *°:ÈUð 1Jñ 1ô 1ñ ˆH�Q˜‘U‰OˆOð
 ˆÝÐ+¨aÑ/Ñ0Ô0ð 	<ð 	<ˆAØ�OŠOÐ-˜DœN¨1Ô-¨h°q¬kÑ:Ô:Ñ;Ô;Ð;Ð;à�Š˜ œÑ%Ô%Ð%åÐ+¨aÑ/°°BÑ7Ô7ð 	ð 	ˆAÝœ-×3Ò3Ø˜” (¨1¤+Ô"3°A°B°BÔ"7¸jÐX]ð 4ñ ô ˆH�Q‰KˆKõ ”9˜X¨1Ð-Ñ-Ô-ˆØ×$Ò$ XÑ.Ô.ˆØ—’ Ñ(Ô(ˆàˆr   )r   r   r   r   r   r&   r©  r(   rl   r·  rj   ro   rp   s   @r   r«  r«  “  s­   ø€ € € € € ðð ð 
Ð3ð  
¸ð  
ð  
ð  
ð  
ð  
ð  
ðD1¨¨e¬lÔ);ð 1ÀÄð 1ð 1ð 1ð 1ð
¨T°%´,Ô-?ð ÀEÄLð ð ð ð ð ð ð ð r   r«  c                   ó†   ‡ — e Zd ZdZ	 	 	 ddedededeeeef         z  d	d
f
ˆ fd„Zdee	j
                 d	e	j
        fd„Zˆ xZS )ÚData2VecVisionFCNHeadaå  
    Fully Convolution Networks for Semantic Segmentation. This head is implemented of
    [FCNNet](https://huggingface.co/papers/1411.4038>).

    Args:
        config (Data2VecVisionConfig): Configuration.
        in_channels
        kernel_size (int): The kernel size for convs in the head. Default: 3.
        dilation (int): The dilation rate for convs in the head. Default: 1.


    Based on OpenMMLab's implementation, found in https://github.com/open-mmlab/mmsegmentation.
    rC   r   r   r"   Úin_indexrt   r‹  r#   Nc           
      óÌ  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        |j        | _        |j        | _	        || _
        |dz  |z  }t          j        ¦   «         | _        | j        dk    r†| j                             t          | j        | j        |||¬¦  «        ¦  «         t!          | j        dz
  ¦  «        D ]8}| j                             t          | j        | j        |||¬¦  «        ¦  «         Œ9| j	        r-t          | j        | j        z   | j        ||dz  ¬¦  «        | _        t          j        | j        |j        d¬¦  «        | _        d S )NrC   r   )rt   rŠ  r‹  r   r®  r˜  )r%   r&   r*   rˆ  Úauxiliary_channelsr–  Úauxiliary_num_convsÚ	num_convsÚauxiliary_concat_inputÚconcat_inputrÂ  r   r;  Úconvsr´  r†  r<  Úconv_catrx   r{  r|  )r<   r"   rÂ  rt   r‹  Úconv_paddingrb   r=   s          €r   r&   zData2VecVisionFCNHead.__init__ô  sz  ø€ õ 	‰Œ×ÒÑÔÐØ!Ô-ˆÔØÔ1ˆŒØÔ3ˆŒØ"Ô9ˆÔØ ˆŒà# qÑ(¨HÑ4ˆÝ”]‘_”_ˆŒ
ØŒ>˜AÒÐØŒJ×ÒÝ'ØÔ$ d¤mÀÐVbÐmuðñ ô ñô ð õ
 ˜4œ>¨AÑ-Ñ.Ô.ð 	ð 	�Ø”
×!Ò!Ý+ØœØœØ$/Ø ,Ø!)ðñ ô ñô ð ð ð Ôð 	Ý3ØÔ  4¤=Ñ0°$´-È[ÐbmÐqrÑbrðñ ô ˆDŒMõ œ) D¤M°6Ô3DÐRSÐTÑTÔTˆŒˆˆr   r¸  c                 óÜ   — || j                  }|}| j        D ]} ||¦  «        }Œ| j        r+|                      t	          j        ||gd¬¦  «        ¦  «        }|                      |¦  «        }|S )Nr   rI   )rÂ  rÉ  rÈ  rÊ  r(   rS   r|  )r<   r¸  Úresidualr—   r›  s        r   rj   zData2VecVisionFCNHead.forward  s~   € Ø(¨¬Ô7ˆØ ˆØ”Jð 	0ð 	0ˆDØ ˜D Ñ/Ô/ˆMˆMØÔð 	WØ ŸMšM­%¬)°X¸}Ð4MÐSTÐ*UÑ*UÔ*UÑVÔVˆMØŸš¨Ñ6Ô6ˆØÐr   )rC   r   r   )r   r   r   r   r   rm   r®   r&   r©  r(   rl   rj   ro   rp   s   @r   rÁ  rÁ  å  sÌ   ø€ € € € € ðð ð" ØØ*+ð%Uð %Uà$ð%Uð ð%Uð ð	%Uð
 ˜˜c 3˜hœÑ'ð%Uð 
ð%Uð %Uð %Uð %Uð %Uð %UðN¨T°%´,Ô-?ð ÀEÄLð ð ð ð ð ð ð ð r   rÁ  c                   ó¨   ‡ — e Zd Zdeddfˆ fd„Zd„ Ze	 	 	 	 	 	 ddej        dz  dej        dz  d	e	dz  d
e	dz  de	de	dz  de
ez  fd„¦   «         Zˆ xZS )Ú%Data2VecVisionForSemanticSegmentationr"   r#   Nc                 óP  •— t          ¦   «                              |¦  «         |j        | _        t          |d¬¦  «        | _        t          | j        j        ¦  «        dk    rt          d¦  «        ‚t          j
        t          j        |j        |j        dd¬¦  «        t          j        |j        ¦  «        t          j        ¦   «         t          j        |j        |j        dd¬¦  «        ¦  «        | _        t          j
        t          j        |j        |j        dd¬¦  «        ¦  «        | _        t          j        ¦   «         | _        t          j        dd¬¦  «        | _        t+          |¦  «        | _        |j        rt1          |¦  «        nd | _        |                      ¦   «          d S )NFrz  r­  zÑData2VecVisionForSemanticSegmentation requires config.out_indices to be a list of 4 integers, specifying which features to use from the backbone. One can use [3, 5, 7, 11] in case of a base-sized architecture.rC   rs   )r%   r&   r{  r`  rN  r°  r"   Úout_indicesr{   r   Ú
SequentialÚConvTranspose2dr*   r�  ÚGELUÚfpn1Úfpn2rÿ   Úfpn3Ú	MaxPool2dÚfpn4r«  Údecode_headÚuse_auxiliary_headrÁ  Úauxiliary_headrh  rÁ   s     €r   r&   z.Data2VecVisionForSemanticSegmentation.__init__)  sy  ø€ Ý‰Œ×Ò˜Ñ Ô Ð à Ô+ˆŒÝ2°6ÈUÐSÑSÔSˆÔõ ˆtŒ{Ô&Ñ'Ô'¨1Ò,Ð,Ýð-ñô ð õ
 ”MÝÔ˜vÔ1°6Ô3EÐSTÐ]^Ð_Ñ_Ô_ÝŒN˜6Ô-Ñ.Ô.ÝŒG‰IŒIÝÔ˜vÔ1°6Ô3EÐSTÐ]^Ð_Ñ_Ô_ñ	
ô 
ˆŒ	õ ”MÝÔ˜vÔ1°6Ô3EÐSTÐ]^Ð_Ñ_Ô_ñ
ô 
ˆŒ	õ ”K‘M”MˆŒ	Ý”L¨Q°qÐ9Ñ9Ô9ˆŒ	õ 2°&Ñ9Ô9ˆÔØ?EÔ?XÐbÕ3°FÑ;Ô;Ð;Ð^bˆÔð 	�ŠÑÔÐÐÐr   c                 óZ  — t           j                             ||j        dd …         dd¬¦  «        }|�0t           j                             ||j        dd …         dd¬¦  «        }t	          | j        j        ¬¦  «        } |||¦  «        }|}|� |||¦  «        }	|| j        j        |	z  z  }|S )Nr›   r   FrE   )Úignore_index)r   rP   rQ   rK   r   r"   Úsemantic_loss_ignore_indexÚauxiliary_loss_weight)
r<   r‚  Úauxiliary_logitsr}  Úupsampled_logitsÚupsampled_auxiliary_logitsÚloss_fctÚ	main_lossr�  Úauxiliary_losss
             r   Úcompute_lossz2Data2VecVisionForSemanticSegmentation.compute_lossI  sÍ   € åœ=×4Ò4Ø˜œ b c cÔ*°È5ð 5ñ 
ô 
Ðð Ð'Ý)+¬×)BÒ)BØ  v¤|°B°C°CÔ'8¸zÐY^ð *Cñ *ô *Ð&õ $°´Ô1WÐXÑXÔXˆØ�HÐ-¨vÑ6Ô6ˆ	ØˆØÐ'Ø%˜XÐ&@À&ÑIÔIˆNØ�D”KÔ5¸ÑFÑFˆDàˆr   Fr[   r}  r˜   r?  rZ   r@  c                 óZ  ‡ ‡‡— |�|n‰ j         j        }|�|n‰ j         j        }|�‰ j         j        dk    rt	          d¦  «        ‚‰                      ||d||¬¦  «        }|r|j        n|d         }	ˆ fd„t          |	¦  «        D ¦   «         }
|j        d         Š‰ j         j	        ‰ j         j
        z  Šˆˆfd„|
D ¦   «         }
‰ j        ‰ j        ‰ j        ‰ j        g}t          t!          |
¦  «        ¦  «        D ]} ||         |
|         ¦  «        |
|<   Œ‰                      |
¦  «        }d}‰ j        �‰                      |
¦  «        }d}|�‰                      |||¦  «        }|s)|r|f|dd…         z   }n|f|d	d…         z   }|�|f|z   n|S t)          |||r|j        nd|j        ¬
¦  «        S )a‘  
        labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*):
            Ground truth semantic segmentation maps for computing the loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels > 1`, a classification loss is computed (Cross-Entropy).

        Examples:

        ```python
        >>> from transformers import AutoImageProcessor, Data2VecVisionForSemanticSegmentation
        >>> from PIL import Image
        >>> import httpx
        >>> from io import BytesIO

        >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
        >>> with httpx.stream("GET", url) as response:
        ...     image = Image.open(BytesIO(response.read()))

        >>> image_processor = AutoImageProcessor.from_pretrained("facebook/data2vec-vision-base")
        >>> model = Data2VecVisionForSemanticSegmentation.from_pretrained("facebook/data2vec-vision-base")

        >>> inputs = image_processor(images=image, return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> # logits are of shape (batch_size, num_labels, height, width)
        >>> logits = outputs.logits
        ```Nr   z/The number of labels should be greater than oneTr  c                 ó<   •— g | ]\  }}|d z   ‰j         j        v ¯|‘ŒS ræ   )r"   rÑ  )r0  ÚidxÚfeaturer<   s      €r   r2  zAData2VecVisionForSemanticSegmentation.forward.<locals>.<listcomp>”  s5   ø€ ÐwÐwÐw¡  WÐTWÐZ[ÑT[Ð_cÔ_jÔ_vÐTvÐTv�GÐTvÐTvÐTvr   r   c                 óˆ   •— g | ]>}|d d …dd …d d …f                               ddd¦  «                             ‰d‰‰¦  «        ‘Œ?S )Nr   r   rC   rB   )rO   rN   )r0  r1  re   Úpatch_resolutions     €€r   r2  zAData2VecVisionForSemanticSegmentation.forward.<locals>.<listcomp>—  sd   ø€ ð 
ð 
ð 
ØijˆAˆaˆaˆa����Q�Q�QˆhŒK×Ò  1 aÑ(Ô(×0Ò0°¸RÐAQÐScÑdÔdð
ð 
ð 
r   rC   r€  )r"   r@  r?  r{  r{   rN  r—   rG  rK   r2   r0   rÕ  rÖ  r×  rÙ  r<  r°  rÚ  rÜ  rç  r   rF  )r<   r[   r}  r˜   r?  rZ   r@  rm  r¬   r¸  ÚfeaturesÚopsr6  r‚  rá  r�  rÏ   re   rí  s   `                @@r   rj   z-Data2VecVisionForSemanticSegmentation.forward\  sA  øøø€ ðH &1Ð%<�k�kÀ$Ä+ÔBYˆà$8Ð$DÐ Ð È$Ì+ÔJjð 	ð Ð $¤+Ô"8¸AÒ"=Ð"=ÝÐNÑOÔOÐOà×&Ò&ØØ/Ø!%Ø%=Ø#ð 'ñ 
ô 
ˆð :EÐ T Ô 5Ð 5È'ÐRSÌ*Ðð xÐwÐwÐwµ	Ð:OÑ0PÔ0PÐwÑwÔwˆØ!Ô'¨Ô*ˆ
Øœ;Ô1°T´[Ô5KÑKÐð
ð 
ð 
ð 
ð 
Ønvð
ñ 
ô 
ˆð
 Œy˜$œ) T¤Y°´	Ð:ˆÝ•s˜8‘}”}Ñ%Ô%ð 	.ð 	.ˆAØ ˜#˜aœ& ¨!¤Ñ-Ô-ˆH�Q‰KˆKà×!Ò! (Ñ+Ô+ˆàÐØÔÐ*Ø#×2Ò2°8Ñ<Ô<ÐàˆØÐØ×$Ò$ VÐ-=¸vÑFÔFˆDàð 	FØ#ð 1Ø ˜ W¨Q¨R¨R¤[Ñ0��à ˜ W¨Q¨R¨R¤[Ñ0�Ø)-Ð)9�T�G˜fÑ$Ð$¸vÐEå&ØØØ3GÐQ˜'Ô/Ð/ÈTØÔ)ð	
ñ 
ô 
ð 	
r   r„  )r   r   r   r   r&   rç  r   r(   rl   r’   r®   r   rj   ro   rp   s   @r   rÏ  rÏ  &  s	  ø€ € € € € ðÐ3ð ¸ð ð ð ð ð ð ð@ð ð ð& ð -1Ø&*Ø)-Ø,0Ø).Ø#'ðY
ð Y
à”l TÑ)ðY
ð ”˜tÑ#ðY
ð   $™;ð	Y
ð
 # T™kðY
ð #'ðY
ð ˜D‘[ðY
ð 
Ð(Ñ	(ðY
ð Y
ð Y
ñ „^ðY
ð Y
ð Y
ð Y
ð Y
r   rÏ  )rx  rÏ  r`  rM  )?r   Úcollections.abcr3   rŸ   Údataclassesr   Útypingr   r(   r   Útorch.nnr   Ú r   rR  Úactivationsr	   Úmodeling_layersr
   Úmodeling_outputsr   r   r   r   Úmodeling_utilsr   Úpytorch_utilsr   Úutilsr   r   r   Úconfiguration_data2vec_visionr   Ú
get_loggerr   r·   r   ÚModuler!   r.   r‚   r°   r¾   rÌ   rÊ   rÔ   rÞ   râ   ró   r”   r,  rM  r`  rf  rx  r†  r”  r   r«  rÁ  rÏ  Ú__all__r   r   r   ú<module>rÿ     s<  ðð $Ð #à Ð Ð Ð Ø €€€Ø !Ð !Ð !Ð !Ð !Ð !Ø Ð Ð Ð Ð Ð à €€€Ø Ð Ð Ð Ð Ð Ø %Ð %Ð %Ð %Ð %Ð %à &Ð &Ð &Ð &Ð &Ð &Ø !Ð !Ð !Ð !Ð !Ð !Ø 9Ð 9Ð 9Ð 9Ð 9Ð 9ðð ð ð ð ð ð ð ð ð ð ð ð .Ð -Ð -Ð -Ð -Ð -Ø @Ð @Ð @Ð @Ð @Ð @Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø ?Ð ?Ð ?Ð ?Ð ?Ð ?ð 
ˆÔ	˜HÑ	%Ô	%€ð €ððñ ô ð
 ðð ð ð ð Ð+Eñ ô ñ „ñô ðð\7ð \7ð \7ð \7ð \7˜rœyñ \7ô \7ð \7ð@#7ð #7ð #7ð #7ð #7 B¤Iñ #7ô #7ð #7ðNFð Fð Fð Fð F "¤)ñ Fô Fð FðT0#ð 0#ð 0#ð 0#ð 0#Ð&Añ 0#ô 0#ð 0#ðhð ð ð ð ˜rœyñ ô ð ð& )Ø+ð*ð *Ð &ðð ð ð ð ˜bœiñ ô ð ð6ð ð ð ð  ¤ñ ô ð ð"
ð 
ð 
ð 
ð 
˜2œ9ñ 
ô 
ð 
ð%ð %ð %ð %ð %˜RœYñ %ô %ð %ð2>ð >ð >ð >ð >Ð4ñ >ô >ð >ðDQ3ð Q3ð Q3ð Q3ð Q3¨¬ñ Q3ô Q3ð Q3ðjE
ð E
ð E
ð E
ð E
˜BœIñ E
ô E
ð E
ðP ðTð Tð Tð Tð T Oñ Tô Tñ „ðTð8 ðD
ð D
ð D
ð D
ð D
Ð7ñ D
ô D
ñ „ðD
ðP	vð 	vð 	vð 	vð 	v˜2œ9ñ 	vô 	vð 	vð €ððñ ô ð8
ð 8
ð 8
ð 8
ð 8
Ð+Hñ 8
ô 8
ñô ð8
ðxð ð ð ð ˜bœiñ ô ð ðF
ð 
ð 
ð 
ð 
¨¬	ñ 
ô 
ð 
ðSð Sð Sð Sð S¨¬ñ Sô Sð Sð>Nð Nð Nð Nð N˜RœYñ Nô Nð Nðd>ð >ð >ð >ð >˜BœIñ >ô >ð >ðB ðN
ð N
ð N
ð N
ð N
Ð,Iñ N
ô N
ñ „ðN
ðbð ð €€€r   