§
    9²¸j.»  ã                   óê  — d 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mZ ddlmZ ddlmZmZmZ ddlmZmZ ddlmZ ddlmZmZm Z m!Z!m"Z" ddl#m$Z$m%Z% ddl&m'Z' ddl(m)Z)  e!j*        e+¦  «        Z, e d¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Z- e d¬¦  «        e G d„ de¦  «        ¦   «         ¦   «         Z. G d„ dej/        ¦  «        Z0 G d„ dej/        ¦  «        Z1 G d„ d ej/        ¦  «        Z2	 	 dXd"ej/        d#ej3        d$ej3        d%ej3        d&ej3        dz  d'e4dz  d(e4d)ee         fd*„Z5 G d+„ d,ej/        ¦  «        Z6 G d-„ d.ej/        ¦  «        Z7 G d/„ d0ej/        ¦  «        Z8 G d1„ d2ej/        ¦  «        Z9 G d3„ d4ej/        ¦  «        Z: G d5„ d6e¦  «        Z; G d7„ d8ej/        ¦  «        Z<d9„ Z= G d:„ d;ej/        ¦  «        Z> G d<„ d=ej/        ¦  «        Z? G d>„ d?ej/        ¦  «        Z@ G d@„ dAej/        ¦  «        ZAe  G dB„ dCe¦  «        ¦   «         ZB G dD„ dEej/        ¦  «        ZCe  G dF„ dGeB¦  «        ¦   «         ZD G dH„ dIej/        ¦  «        ZE G dJ„ dKej/        ¦  «        ZF G dL„ dMej/        ¦  «        ZG e dN¬¦  «         G dO„ dPeB¦  «        ¦   «         ZH G dQ„ dRej/        ¦  «        ZI G dS„ dTej/        ¦  «        ZJe  G dU„ dVeB¦  «        ¦   «         ZKg dW¢ZLdS )YzèPyTorch DPT (Dense Prediction Transformers) model.

This implementation is heavily inspired by OpenMMLab's implementation, found here:
https://github.com/open-mmlab/mmsegmentation/blob/master/mmseg/models/decode_heads/dpt_head.py.

é    N)ÚCallable)Ú	dataclass)Únn)ÚCrossEntropyLossé   )Úinitialization)ÚACT2FN)Úfilter_output_hidden_statesÚload_backbone)ÚGradientCheckpointingLayer)ÚBaseModelOutputÚDepthEstimatorOutputÚSemanticSegmenterOutput)ÚALL_ATTENTION_FUNCTIONSÚPreTrainedModel)ÚUnpack)ÚModelOutputÚTransformersKwargsÚauto_docstringÚloggingÚ	torch_int)Úcan_return_tupleÚmerge_with_config_defaults)Úcapture_outputsé   )Ú	DPTConfigz£
    Base class for model's outputs that also contains intermediate activations that can be used at later stages. Useful
    in the context of Vision models.:
    )Úcustom_introc                   ó`   — e Zd ZU dZdZej        dz  ed<   dZe	ej        df         dz  ed<   dS )Ú*BaseModelOutputWithIntermediateActivationsak  
    last_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
        Sequence of hidden-states at the output of the last layer of the model.
    intermediate_activations (`tuple(torch.FloatTensor)`, *optional*):
        Intermediate activations that can be used to compute hidden states of the model at various layers.
    NÚlast_hidden_states.Úintermediate_activations)
Ú__name__Ú
__module__Ú__qualname__Ú__doc__r    ÚtorchÚFloatTensorÚ__annotations__r!   Útuple© ó    út/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/dpt/modeling_dpt.pyr   r   -   sZ   € € € € € € ðð ð 48Ð˜Ô)¨DÑ0Ð7Ð7Ñ7ØEIÐ˜e EÔ$5°sÐ$:Ô;¸dÑBÐIÐIÑIÐIÐIr+   r   z¶
    Base class for model's outputs that also contains a pooling of the last hidden states as well as intermediate
    activations that can be used by the model at later stages.
    c                   óÚ   — e Zd ZU dZdZej        dz  ed<   dZej        dz  ed<   dZ	e
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dZe
ej        df         dz  ed<   dS )	Ú4BaseModelOutputWithPoolingAndIntermediateActivationsaÐ  
    pooler_output (`torch.FloatTensor` of shape `(batch_size, hidden_size)`):
        Last layer hidden-state of the first token of the sequence (classification token) after further processing
        through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
        the classification token after processing through a linear layer and a tanh activation function. The linear
        layer weights are trained from the next sentence prediction (classification) objective during pretraining.
    intermediate_activations (`tuple(torch.FloatTensor)`, *optional*):
        Intermediate activations that can be used to compute hidden states of the model at various layers.
    NÚlast_hidden_stateÚpooler_output.Úhidden_statesÚ
attentionsr!   )r"   r#   r$   r%   r/   r&   r'   r(   r0   r1   r)   r2   r!   r*   r+   r,   r.   r.   @   s¶   € € € € € € ðð ð 37Ð�uÔ(¨4Ñ/Ð6Ð6Ñ6Ø.2€M�5Ô$ tÑ+Ð2Ð2Ñ2Ø:>€M�5˜Ô*¨CÐ/Ô0°4Ñ7Ð>Ð>Ñ>Ø7;€J��eÔ'¨Ð,Ô-°Ñ4Ð;Ð;Ñ;ØEIÐ˜e EÔ$5°sÐ$:Ô;¸dÑBÐIÐIÑIÐIÐIr+   r.   c                   ón   ‡ — e Zd ZdZddedeeef         dz  fˆ fd„Zdd„Z	 dd	e	j
        d
edefd„Zˆ xZS )ÚDPTViTHybridEmbeddingszì
    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.
    NÚconfigÚfeature_sizec                 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  }t          |¦  «        | _        | j        j        d         }t          | j        j        ¦  «        dk    r)t          dt          | j        j        ¦  «        › �¦  «        ‚ddg| _        |€|j        }	|	dd …         }|	d         }n7t          |t          j        j	        ¦  «        r|n||f}| j        j        d         }|| _        |d         | _        || _        t#          j        ||d¬¦  «        | _        t#          j        t+          j        dd|j        ¦  «        ¦  «        | _        t#          j        t+          j        d|dz   |j        ¦  «        ¦  «        | _        d S )Nr   r   éÿÿÿÿr   z1Expected backbone to have 3 output features, got éþÿÿÿ©Úkernel_size)ÚsuperÚ__init__Ú
image_sizeÚ
patch_sizeÚnum_channelsÚhidden_sizeÚ
isinstanceÚcollectionsÚabcÚIterabler   ÚbackboneÚchannelsÚlenÚ
ValueErrorÚresidual_feature_map_indexÚbackbone_featmap_shaper   ÚConv2dÚ
projectionÚ	Parameterr&   ÚzerosÚ	cls_tokenÚposition_embeddings)Úselfr5   r6   r>   r?   r@   rA   Únum_patchesÚfeature_dimÚfeat_map_shapeÚ	__class__s             €r,   r=   zDPTViTHybridEmbeddings.__init__`   sñ  ø€ Ý‰Œ×ÒÑÔÐØ!'Ô!2°FÔ4E�Jˆ
Ø$*Ô$7¸Ô9K�kˆå#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ý#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ø! !”}¨
°1¬Ñ5¸*ÀQ¼-È:ÐVWÌ=Ñ:XÑYˆå% fÑ-Ô-ˆŒØ”mÔ,¨RÔ0ˆÝˆtŒ}Ô%Ñ&Ô&¨!Ò+Ð+ÝÐnÕQTÐUYÔUbÔUkÑQlÔQlÐnÐnÑoÔoÐoØ+,¨a¨&ˆÔ'àÐØ#Ô:ˆNØ)¨"¨#¨#Ô.ˆLØ(¨Ô+ˆKˆKõ !+¨<½¼Ô9QÑ RÔ RÐt��ÐYeÐgsÐXtð ð œ-Ô0°Ô4ˆKà$ˆŒØ$ Qœ-ˆŒØ(ˆÔåœ) K°È!ÐLÑLÔLˆŒåœ¥e¤k°!°Q¸Ô8JÑ&KÔ&KÑLÔLˆŒÝ#%¤<µ´¸A¸{ÈQ¹ÐPVÔPbÑ0cÔ0cÑ#dÔ#dˆÔ Ð Ð r+   r   c                 ó¨  — |d d …d |…f         }|d|d …f         }t          |j        d         dz  ¦  «        }|                     d||d¦  «                             dddd¦  «        }t          j                             |||fd¬¦  «        }|                     dddd¦  «                             d||z  d¦  «        }t          j        ||gd¬	¦  «        }|S ©
Nr   ç      à?r   r8   r   é   Úbilinear)ÚsizeÚmode©Údim)	r   ÚshapeÚreshapeÚpermuter   Ú
functionalÚinterpolater&   Úcat©rR   ÚposembÚgrid_size_heightÚgrid_size_widthÚstart_indexÚ
posemb_tokÚposemb_gridÚold_grid_sizes           r,   Ú_resize_pos_embedz(DPTViTHybridEmbeddings._resize_pos_embed‚   sê   € Ø˜A˜A˜A˜| ˜|˜OÔ,ˆ
Ø˜Q   ˜_Ô-ˆå! +Ô"3°AÔ"6¸#Ñ"=Ñ>Ô>ˆà!×)Ò)¨!¨]¸MÈ2ÑNÔN×VÒVÐWXÐZ[Ð]^Ð`aÑbÔbˆÝ”m×/Ò/°ÐCSÐUdÐBeÐlvÐ/ÑwÔwˆØ!×)Ò)¨!¨Q°°1Ñ5Ô5×=Ò=¸aÐAQÐTcÑAcÐegÑhÔhˆå”˜J¨Ð4¸!Ð<Ñ<Ô<ˆàˆr+   FÚpixel_valuesÚinterpolate_pos_encodingÚreturnc                 óÊ  ‡— |j         \  }}}}|| j        k    rt          d¦  «        ‚|sT|| j        d         k    s|| j        d         k    r2t          d|› d|› d| j        d         › d| j        d         › d�	¦  «        ‚|                      | j        || j        z  || j        z  ¦  «        }|                      |¦  «        Š‰j        d         }ˆfd	„| j	        D ¦   «         }	|  
                    |¦  «                             d
¦  «                             dd
¦  «        }
| j                             |dd¦  «        }t          j        ||
fd¬¦  «        }
|
|z   }
t#          |
|	¬¦  «        S )NúeMake sure that the channel dimension of the pixel values match with the one set in the configuration.r   r   zInput image size (Ú*z) doesn't match model (z).r8   c                 ó*   •— g | ]}‰j         |         ‘ŒS r*   )Úfeature_maps)Ú.0ÚindexÚbackbone_outputs     €r,   ú
<listcomp>z2DPTViTHybridEmbeddings.forward.<locals>.<listcomp>¨   s!   ø€ ÐqÐqÐqÈ Ô <¸UÔ CÐqÐqÐqr+   rZ   r^   )r    r!   )r`   r@   rI   r>   rn   rQ   r?   rF   rv   rJ   rM   ÚflattenÚ	transposerP   Úexpandr&   re   r   )rR   ro   rp   Ú
batch_sizer@   ÚheightÚwidthrQ   ÚfeaturesÚoutput_hidden_statesÚ
embeddingsÚ
cls_tokensry   s               @r,   ÚforwardzDPTViTHybridEmbeddings.forward�   sÊ  ø€ ð 3?Ô2DÑ/ˆ
�L &¨%Ø˜4Ô,Ò,Ð,ÝØwñô ð ð (ð 	Ø˜œ¨Ô+Ò+Ð+¨u¸¼ÈÔ8JÒ/JÐ/JÝ ðE¨ð Eð E°%ð Eð EØœ¨Ô+ðEð EØ.2¬o¸aÔ.@ðEð Eð Eñô ð ð
 #×4Ò4ØÔ$ f°´Ñ&?ÀÈ$Ì/ÑAYñ
ô 
Ðð Ÿ-š-¨Ñ5Ô5ˆà"Ô/°Ô3ˆð  rÐqÐqÐqÐQUÔQpÐqÑqÔqÐà—_’_ XÑ.Ô.×6Ò6°qÑ9Ô9×CÒCÀAÀqÑIÔIˆ
à”^×*Ò*¨:°r¸2Ñ>Ô>ˆ
Ý”Y 
¨JÐ7¸QÐ?Ñ?Ô?ˆ
ð  Ð"5Ñ5ˆ
õ :Ø)Ø%9ð
ñ 
ô 
ð 	
r+   ©N©r   ©F)r"   r#   r$   r%   r   r)   Úintr=   rn   r&   ÚTensorÚboolr   r…   Ú__classcell__©rV   s   @r,   r4   r4   Y   s¿   ø€ € € € € ðð ð eð  e˜yð  e¸¸cÀ3¸h¼È$Ñ8Nð  eð  eð  eð  eð  eð  eðDð ð ð ð LQð&
ð &
Ø!œLð&
ØDHð&
à	3ð&
ð &
ð &
ð &
ð &
ð &
ð &
ð &
r+   r4   c                   óD   ‡ — e Zd ZdZˆ fd„Zdd„Zdej        defd„Z	ˆ xZ
S )	ÚDPTViTEmbeddingszB
    Construct the CLS token, position and patch embeddings.

    c                 ó   •— t          ¦   «                              ¦   «          t          j        t	          j        dd|j        ¦  «        ¦  «        | _        t          |¦  «        | _	        | j	        j
        }t          j        t	          j        d|dz   |j        ¦  «        ¦  «        | _        t          j        |j        ¦  «        | _        || _        d S )Nr   )r<   r=   r   rN   r&   rO   rA   rP   ÚDPTViTPatchEmbeddingsÚpatch_embeddingsrS   rQ   ÚDropoutÚhidden_dropout_probÚdropoutr5   )rR   r5   rS   rV   s      €r,   r=   zDPTViTEmbeddings.__init__¿   s�   ø€ Ý‰Œ×ÒÑÔÐåœ¥e¤k°!°Q¸Ô8JÑ&KÔ&KÑLÔLˆŒÝ 5°fÑ =Ô =ˆÔØÔ+Ô7ˆÝ#%¤<µ´¸A¸{ÈQ¹ÐPVÔPbÑ0cÔ0cÑ#dÔ#dˆÔ Ý”z &Ô"<Ñ=Ô=ˆŒØˆŒˆˆr+   r   c                 ó¸  — |d d …d |…f         }|d|d …f         }t          |                     d¦  «        dz  ¦  «        }|                     d||d¦  «                             dddd¦  «        }t          j                             |||fd¬¦  «        }|                     dddd¦  «                             d||z  d¦  «        }t          j        ||gd¬	¦  «        }|S rX   )	r   r\   ra   rb   r   rc   rd   r&   re   rf   s           r,   rn   z"DPTViTEmbeddings._resize_pos_embedÉ   sð   € Ø˜A˜A˜A˜| ˜|˜OÔ,ˆ
Ø˜Q   ˜_Ô-ˆå! +×"2Ò"2°1Ñ"5Ô"5¸Ñ"<Ñ=Ô=ˆà!×)Ò)¨!¨]¸MÈ2ÑNÔN×VÒVÐWXÐZ[Ð]^Ð`aÑbÔbˆÝ”m×/Ò/°ÐCSÐUdÐBeÐlvÐ/ÑwÔwˆØ!×)Ò)¨!¨Q°°1Ñ5Ô5×=Ò=¸aÐAQÐTcÑAcÐegÑhÔhˆå”˜J¨Ð4¸!Ð<Ñ<Ô<ˆàˆr+   ro   rq   c                 óŒ  — |j         \  }}}}| j        j        }|                      | j        ||z  ||z  ¦  «        }|                      |¦  «        }|                     ¦   «         \  }}	}
| j                             |dd¦  «        }t          j
        ||fd¬¦  «        }||z   }|                      |¦  «        }t          |¬¦  «        S )Nr8   r   r^   )r    )r`   r5   r?   rn   rQ   r’   r\   rP   r}   r&   re   r•   r   )rR   ro   r~   r@   r   r€   r?   rQ   rƒ   Úseq_lenÚ_r„   s               r,   r…   zDPTViTEmbeddings.forward×   s×   € Ø2>Ô2DÑ/ˆ
�L &¨%ð ”[Ô+ˆ
Ø"×4Ò4ØÔ$ f°
Ñ&:¸EÀZÑ<Oñ
ô 
Ðð ×*Ò*¨<Ñ8Ô8ˆ
à!+§¢Ñ!2Ô!2Ñˆ
�G˜Qð ”^×*Ò*¨:°r¸2Ñ>Ô>ˆ
Ý”Y 
¨JÐ7¸QÐ?Ñ?Ô?ˆ
ð  Ð"5Ñ5ˆ
à—\’\ *Ñ-Ô-ˆ
å9ÈZÐXÑXÔXÐXr+   r‡   )r"   r#   r$   r%   r=   rn   r&   rŠ   r   r…   rŒ   r�   s   @r,   r�   r�   ¹   s…   ø€ € € € € ðð ð
ð ð ð ð ðð ð ð ðY E¤Lð YÐ5_ð Yð Yð Yð Yð Yð Yð Yð Yr+   r�   c                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Zˆ xZ	S )r‘   z$
    Image to Patch Embedding.

    r5   c                 óÌ  •— 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  }|| _        || _        || _        || _
        t          j        ||||¬¦  «        | _        d S )Nr   r   )r;   Ústride)r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   rS   r   rL   rM   )rR   r5   r>   r?   r@   rA   rS   rV   s          €r,   r=   zDPTViTPatchEmbeddings.__init__ö   sá   ø€ Ý‰Œ×ÒÑÔÐØ!'Ô!2°FÔ4E�Jˆ
Ø$*Ô$7¸Ô9K�kˆå#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ý#-¨j½+¼/Ô:RÑ#SÔ#SÐq�Z�ZÐZdÐfpÐYqˆ
Ø! !”}¨
°1¬Ñ5¸*ÀQ¼-È:ÐVWÌ=Ñ:XÑYˆØ$ˆŒØ$ˆŒØ(ˆÔØ&ˆÔåœ) L°+È:Ð^hÐiÑiÔiˆŒˆˆr+   ro   rq   c                 óÊ   — |j         \  }}}}|| j        k    rt          d¦  «        ‚|                      |¦  «                             d¦  «                             dd¦  «        }|S )Nrs   rZ   r   )r`   r@   rI   rM   r{   r|   )rR   ro   r~   r@   r   r€   rƒ   s          r,   r…   zDPTViTPatchEmbeddings.forward  sm   € Ø2>Ô2DÑ/ˆ
�L &¨%Ø˜4Ô,Ò,Ð,ÝØwñô ð ð —_’_ \Ñ2Ô2×:Ò:¸1Ñ=Ô=×GÒGÈÈ1ÑMÔMˆ
ØÐr+   ©
r"   r#   r$   r%   r   r=   r&   rŠ   r…   rŒ   r�   s   @r,   r‘   r‘   ð   s{   ø€ € € € € ðð ð
j˜yð jð jð jð jð jð jð E¤Lð °U´\ð ð ð ð ð ð ð ð r+   r‘   ç        ÚmoduleÚqueryÚkeyÚvalueÚattention_maskÚscalingr•   Úkwargsc                 ó®  — |€|                      d¦  «        dz  }t          j        ||                     dd¦  «        ¦  «        |z  }|�||z   }t          j                             |d¬¦  «        }t          j                             ||| j        ¬¦  «        }t          j        ||¦  «        }	|	                     dd¦  «         	                    ¦   «         }	|	|fS )Nr8   ç      à¿rZ   r   r^   )ÚpÚtrainingr   )
r\   r&   Úmatmulr|   r   rc   Úsoftmaxr•   rª   Ú
contiguous)
r    r¡   r¢   r£   r¤   r¥   r•   r¦   Úattn_weightsÚattn_outputs
             r,   Úeager_attention_forwardr°     sÈ   € ð €Ø—*’*˜R‘.”. DÑ(ˆõ ”<  s§}¢}°Q¸Ñ':Ô':Ñ;Ô;¸gÑE€LàÐ!Ø# nÑ4ˆå”=×(Ò(¨¸2Ð(Ñ>Ô>€LÝ”=×(Ò(¨¸È6Ì?Ð(Ñ[Ô[€Lå”,˜|¨UÑ3Ô3€KØ×'Ò'¨¨1Ñ-Ô-×8Ò8Ñ:Ô:€Kà˜Ð$Ð$r+   c                   ór   ‡ — e Zd Zdefˆ fd„Zdej        dee         de	ej        ej        f         fd„Z
ˆ xZS )ÚDPTSelfAttentionr5   c                 ó¤  •— t          ¦   «                              ¦   «          |j        |j        z  dk    r0t	          |d¦  «        s t          d|j        › d|j        › d�¦  «        ‚|| _        |j        | _        t          |j        |j        z  ¦  «        | _        | j        | j        z  | _	        |j
        | _        | j        dz  | _        d| _        t          j        |j        | j	        |j        ¬¦  «        | _        t          j        |j        | j	        |j        ¬¦  «        | _        t          j        |j        | j	        |j        ¬¦  «        | _        d S )	Nr   Úembedding_sizezThe hidden size z4 is not a multiple of the number of attention heads ú.r¨   F)Úbias)r<   r=   rA   Únum_attention_headsÚhasattrrI   r5   r‰   Úattention_head_sizeÚall_head_sizeÚattention_probs_dropout_probÚdropout_probr¥   Ú	is_causalr   ÚLinearÚqkv_biasr¡   r¢   r£   ©rR   r5   rV   s     €r,   r=   zDPTSelfAttention.__init__.  sB  ø€ Ý‰Œ×ÒÑÔÐØÔ Ô :Ñ:¸aÒ?Ð?ÍÐPVÐXhÑHiÔHiÐ?Ýð7 6Ô#5ð 7ð 7ØÔ3ð7ð 7ð 7ñô ð ð
 ˆŒØ#)Ô#=ˆÔ Ý#& vÔ'9¸FÔ<VÑ'VÑ#WÔ#WˆÔ Ø!Ô5¸Ô8PÑPˆÔØ"Ô?ˆÔØÔ/°Ñ5ˆŒØˆŒå”Y˜vÔ1°4Ô3EÈFÌOÐ\Ñ\Ô\ˆŒ
Ý”9˜VÔ/°Ô1CÈ&Ì/ÐZÑZÔZˆŒÝ”Y˜vÔ1°4Ô3EÈFÌOÐ\Ñ\Ô\ˆŒ
ˆ
ˆ
r+   r1   r¦   rq   c                 ó~  — |j         d         }|d| j        | j        f} |                      |¦  «        j        |Ž                      dd¦  «        } |                      |¦  «        j        |Ž                      dd¦  «        } |                      |¦  «        j        |Ž                      dd¦  «        }t          j	        | j
        j        t          ¦  «        } || |||d f| j        | j        | j        sdn| j        dœ|¤Ž\  }	}
|	                     ¦   «         d d…         | j        fz   }|	                     |¦  «        }	|	|
fS )Nr   r8   r   rZ   rŸ   )r½   r¥   r•   r9   )r`   r·   r¹   r¢   Úviewr|   r£   r¡   r   Úget_interfacer5   Ú_attn_implementationr°   r½   r¥   rª   r¼   r\   rº   ra   )rR   r1   r¦   r~   Ú	new_shapeÚ	key_layerÚvalue_layerÚquery_layerÚattention_interfaceÚcontext_layerÚattention_probsÚnew_context_layer_shapes               r,   r…   zDPTSelfAttention.forwardB  sd  € ð
 #Ô(¨Ô+ˆ
Ø  DÔ$<¸dÔ>VÐVˆ	à0�D—H’H˜]Ñ+Ô+Ô0°)Ð<×FÒFÀqÈ!ÑLÔLˆ	Ø4�d—j’j Ñ/Ô/Ô4°iÐ@×JÒJÈ1ÈaÑPÔPˆØ4�d—j’j Ñ/Ô/Ô4°iÐ@×JÒJÈ1ÈaÑPÔPˆå(?Ô(MØŒKÔ,Õ.Eñ)
ô )
Ðð *=Ð)<ØØØØØð
*
ð ”nØ”LØ#œ}ÐC�C�C°$Ô2Cð
*
ð 
*
ð ð
*
ð 
*
Ñ&ˆ�ð #0×"4Ò"4Ñ"6Ô"6°s¸°sÔ";¸tÔ?QÐ>SÑ"SÐØ%×-Ò-Ð.EÑFÔFˆà˜oÐ-Ð-r+   )r"   r#   r$   r   r=   r&   rŠ   r   r   r)   r…   rŒ   r�   s   @r,   r²   r²   -  s‘   ø€ € € € € ð]˜yð ]ð ]ð ]ð ]ð ]ð ]ð(.à”|ð.ð Ð+Ô,ð.ð 
ˆuŒ|˜Uœ\Ð)Ô	*ð	.ð .ð .ð .ð .ð .ð .ð .r+   r²   c                   óZ   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        dej        fd„Zˆ xZ	S )ÚDPTViTSelfOutputz 
    The residual connection is defined in ViTLayer instead of here (as is the case with other models), due to the
    layernorm applied before each block.
    r5   c                 óÌ   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          j        |j        ¦  «        | _        d S r†   )	r<   r=   r   r¾   rA   Údenser“   r”   r•   rÀ   s     €r,   r=   zDPTViTSelfOutput.__init__k  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+   r1   Úinput_tensorrq   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r†   ©rÐ   r•   ©rR   r1   rÑ   s      r,   r…   zDPTViTSelfOutput.forwardp  s*   € ØŸ
š
 =Ñ1Ô1ˆØŸš ]Ñ3Ô3ˆØÐr+   rž   r�   s   @r,   rÎ   rÎ   e  s   ø€ € € € € ðð ð
>˜yð >ð >ð >ð >ð >ð >ð
 U¤\ð ÀÄð ÐRWÔR^ð ð ð ð ð ð ð ð r+   rÎ   c                   óX   ‡ — e Zd Zdefˆ fd„Zdej        dee         dej        fd„Z	ˆ xZ
S )ÚDPTViTAttentionr5   c                 ó˜   •— t          ¦   «                              ¦   «          t          |¦  «        | _        t	          |¦  «        | _        d S r†   )r<   r=   r²   Ú	attentionrÎ   ÚoutputrÀ   s     €r,   r=   zDPTViTAttention.__init__x  s;   ø€ Ý‰Œ×ÒÑÔÐÝ)¨&Ñ1Ô1ˆŒÝ& vÑ.Ô.ˆŒˆˆr+   r1   r¦   rq   c                 óT   —  | j         |fi |¤Ž\  }}|                      ||¦  «        }|S r†   )rØ   rÙ   )rR   r1   r¦   Úself_attn_outputr™   rÙ   s         r,   r…   zDPTViTAttention.forward}  s<   € ð
 -˜dœn¨]ÐEÐE¸fÐEÐEÑÐ˜!Ø—’Ð-¨}Ñ=Ô=ˆØˆr+   )r"   r#   r$   r   r=   r&   rŠ   r   r   r…   rŒ   r�   s   @r,   rÖ   rÖ   w  s~   ø€ € € € € ð/˜yð /ð /ð /ð /ð /ð /ð
à”|ðð Ð+Ô,ðð 
Œð	ð ð ð ð ð ð ð r+   rÖ   c                   óH   ‡ — e Zd Zdefˆ fd„Zdej        dej        fd„Zˆ xZS )ÚDPTViTIntermediater5   c                 ó  •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          |j        t          ¦  «        rt          |j                 | _        d S |j        | _        d S r†   )r<   r=   r   r¾   rA   Úintermediate_sizerÐ   rB   Ú
hidden_actÚstrr	   Úintermediate_act_fnrÀ   s     €r,   r=   zDPTViTIntermediate.__init__‰  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$r+   r1   rq   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r†   )rÐ   râ   )rR   r1   s     r,   r…   zDPTViTIntermediate.forward‘  s,   € ØŸ
š
 =Ñ1Ô1ˆØ×0Ò0°Ñ?Ô?ˆØÐr+   ©	r"   r#   r$   r   r=   r&   rŠ   r…   rŒ   r�   s   @r,   rÝ   rÝ   ˆ  sj   ø€ € € € € ð9˜yð 9ð 9ð 9ð 9ð 9ð 9ð U¤\ð °e´lð ð ð ð ð ð ð ð r+   rÝ   c                   óV   ‡ — e Zd Zdefˆ fd„Zdej        dej        dej        fd„Zˆ xZS )ÚDPTViTOutputr5   c                 óÌ   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          j        |j        ¦  «        | _	        d S r†   )
r<   r=   r   r¾   rß   rA   rÐ   r“   r”   r•   rÀ   s     €r,   r=   zDPTViTOutput.__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+   r1   rÑ   rq   c                 ód   — |                       |¦  «        }|                      |¦  «        }||z   }|S r†   rÓ   rÔ   s      r,   r…   zDPTViTOutput.forwardž  s4   € ØŸ
š
 =Ñ1Ô1ˆØŸš ]Ñ3Ô3ˆØ%¨Ñ4ˆØÐr+   rä   r�   s   @r,   ræ   ræ   ˜  su   ø€ € € € € ð>˜yð >ð >ð >ð >ð >ð >ð
 U¤\ð ÀÄð ÐRWÔR^ð ð ð ð ð ð ð ð r+   ræ   c                   ó\   ‡ — e Zd ZdZdefˆ fd„Zdej        dee	         dej        fd„Z
ˆ xZS )ÚDPTViTLayerz?This corresponds to the Block class in the timm implementation.r5   c                 óz  •— t          ¦   «                              ¦   «          |j        | _        d| _        t	          |¦  «        | _        t          |¦  «        | _        t          |¦  «        | _	        t          j        |j        |j        ¬¦  «        | _        t          j        |j        |j        ¬¦  «        | _        d S )Nr   ©Úeps)r<   r=   Úchunk_size_feed_forwardÚseq_len_dimrÖ   rØ   rÝ   Úintermediateræ   rÙ   r   Ú	LayerNormrA   Úlayer_norm_epsÚlayernorm_beforeÚlayernorm_afterrÀ   s     €r,   r=   zDPTViTLayer.__init__©  sš   ø€ Ý‰Œ×ÒÑÔÐØ'-Ô'EˆÔ$ØˆÔÝ(¨Ñ0Ô0ˆŒÝ.¨vÑ6Ô6ˆÔÝ" 6Ñ*Ô*ˆŒÝ "¤¨VÔ-?ÀVÔEZÐ [Ñ [Ô [ˆÔÝ!œ|¨FÔ,>ÀFÔDYÐZÑZÔZˆÔÐÐr+   r1   r¦   rq   c                 óÖ   — |                       |¦  «        } | j        |fi |¤Ž}||z   }|                      |¦  «        }|                      |¦  «        }|                      ||¦  «        }|S r†   )ró   rØ   rô   rð   rÙ   )rR   r1   r¦   Úhidden_states_normÚattention_outputÚlayer_outputs         r,   r…   zDPTViTLayer.forward³  s‚   € ð
 "×2Ò2°=ÑAÔAÐØ)˜4œ>Ð*<ÐGÐGÀÐGÐGÐð )¨=Ñ8ˆð ×+Ò+¨MÑ:Ô:ˆØ×(Ò(¨Ñ6Ô6ˆð —{’{ <°Ñ?Ô?ˆàÐr+   )r"   r#   r$   r%   r   r=   r&   rŠ   r   r   r…   rŒ   r�   s   @r,   rê   rê   ¦  s‹   ø€ € € € € ØIÐIð[˜yð [ð [ð [ð [ð [ð [ðà”|ðð Ð+Ô,ðð 
Œð	ð ð ð ð ð ð ð r+   rê   c                   ól   ‡ — e Zd ZdZˆ fd„Zd„ Zd„ Zd	deej	                 deej	                 fd„Z
ˆ xZS )
ÚDPTReassembleStagea@  
    This class reassembles the hidden states of the backbone into image-like feature representations at various
    resolutions.

    This happens in 3 stages:
    1. Map the N + 1 tokens to a set of N tokens, by taking into account the readout ([CLS]) token according to
       `config.readout_type`.
    2. Project the channel dimension of the hidden states according to `config.neck_hidden_sizes`.
    3. Resizing the spatial dimensions (height, width).

    Args:
        config (`[DPTConfig]`):
            Model configuration class defining the model architecture.
    c                 ó  •— t          ¦   «                              ¦   «          || _        t          j        ¦   «         | _        |j        r|                      |¦  «         n|                      |¦  «         |j	        | _	        d S r†   )
r<   r=   r5   r   Ú
ModuleListÚlayersÚ	is_hybridÚ_init_reassemble_dpt_hybridÚ_init_reassemble_dptÚneck_ignore_stagesrÀ   s     €r,   r=   zDPTReassembleStage.__init__Ø  st   ø€ Ý‰Œ×ÒÑÔÐàˆŒÝ”m‘o”oˆŒØÔð 	.Ø×,Ò,¨VÑ4Ô4Ð4Ð4à×%Ò% fÑ-Ô-Ð-à"(Ô";ˆÔÐÐr+   c           	      ój  — t          t          t          |j        ¦  «        ¦  «        |j        ¦  «        D ]r\  }}|dk    r,| j                             t          j        ¦   «         ¦  «         Œ7|dk    r5| j                             t          ||j        |         |¬¦  «        ¦  «         Œs|j
        dk    rt          d|j
        › d�¦  «        ‚t          j        ¦   «         | _        t          |¦  «        }t          t          |j        ¦  «        ¦  «        D ]Ÿ}|dk    r>| j                             t          j        t          j        ¦   «         ¦  «        ¦  «         ŒF|dk    rS| j                             t          j        t          j        d|z  |¦  «        t"          |j                 ¦  «        ¦  «         Œ dS )a   "
        For DPT-Hybrid the first 2 reassemble layers are set to `nn.Identity()`, please check the official
        implementation: https://github.com/isl-org/DPT/blob/f43ef9e08d70a752195028a51be5e1aff227b913/dpt/vit.py#L438
        for more details.
        r   ©rG   ÚfactorÚprojectzReadout type z! is not supported for DPT-Hybrid.rZ   N)ÚzipÚrangerH   Úneck_hidden_sizesÚreassemble_factorsrý   Úappendr   ÚIdentityÚDPTReassembleLayerÚreadout_typerI   rü   Úreadout_projectsÚ_get_backbone_hidden_sizeÚ
Sequentialr¾   r	   rà   )rR   r5   Úir  rA   s        r,   rÿ   z.DPTReassembleStage._init_reassemble_dpt_hybridä  s�  € õ �U¥3 vÔ'?Ñ#@Ô#@ÑAÔAÀ6ÔC\Ñ]Ô]ð 	tð 	t‰IˆAˆvØ�AŠvˆvØ”×"Ò"¥2¤;¡=¤=Ñ1Ô1Ð1Ð1Ø�Q’�Ø”×"Ò"Õ#5°fÀvÔG_Ð`aÔGbÐkqÐ#rÑ#rÔ#rÑsÔsÐsøàÔ )Ò+Ð+ÝÐc¨VÔ-@ÐcÐcÐcÑdÔdÐdõ !#¤¡¤ˆÔÝ/°Ñ7Ô7ˆÝ•s˜6Ô3Ñ4Ô4Ñ5Ô5ð 	ð 	ˆAØ�AŠvˆvØÔ%×,Ò,­R¬]½2¼;¹=¼=Ñ-IÔ-IÑJÔJÐJÐJØ�Q’�ØÔ%×,Ò,Ý”M¥"¤)¨A°©O¸[Ñ"IÔ"IÍ6ÐRXÔRcÔKdÑeÔeñô ð øð		ð 	r+   c           	      ó:  — t          t          t          |j        ¦  «        ¦  «        |j        ¦  «        D ]:\  }}| j                             t          ||j        |         |¬¦  «        ¦  «         Œ;|j        dk    ržt          j
        ¦   «         | _        t          |¦  «        }t          t          |j        ¦  «        ¦  «        D ]W}| j                             t          j        t          j        d|z  |¦  «        t          |j                 ¦  «        ¦  «         ŒVd S d S )Nr  r  rZ   )r  r  rH   r  r	  rý   r
  r  r  r   rü   r  r  r  r¾   r	   rà   )rR   r5   r  r  rA   r™   s         r,   r   z'DPTReassembleStage._init_reassemble_dptþ  s  € Ý�U¥3 vÔ'?Ñ#@Ô#@ÑAÔAÀ6ÔC\Ñ]Ô]ð 	pð 	p‰IˆAˆvØŒK×ÒÕ1°&À6ÔC[Ð\]ÔC^ÐgmÐnÑnÔnÑoÔoÐoÐoàÔ )Ò+Ð+Ý$&¤M¡O¤OˆDÔ!Ý3°FÑ;Ô;ˆKÝ�3˜vÔ7Ñ8Ô8Ñ9Ô9ð ð �ØÔ%×,Ò,Ý”M¥"¤)¨A°©O¸[Ñ"IÔ"IÍ6ÐRXÔRcÔKdÑeÔeñô ð ð ð	 ,Ð+ðð r+   Nr1   rq   c                 óè  — g }t          |¦  «        D �]Þ\  }}|| j        v�r¹|dd…df         |dd…dd…f         }}|j        \  }}	}
|�|�|                     ||||
¦  «        }n*t	          |	dz  ¦  «        }|                     ||||
¦  «        }|                     dddd¦  «                             ¦   «         }|j        }| j        j        dk    r¦| 	                    d¦  «                             d¦  «        }| 
                    d¦  «                             |¦  «        } | j        |         t          j        ||fd	¦  «        ¦  «        }|                     ddd¦  «                             |¦  «        }nP| j        j        d
k    r@| 	                    d¦  «        | 
                    d	¦  «        z   }|                     |¦  «        } | j        |         |¦  «        }|                     |¦  «         �Œà|S )zÇ
        Args:
            hidden_states (`list[torch.FloatTensor]`, each of shape `(batch_size, sequence_length + 1, hidden_size)`):
                List of hidden states from the backbone.
        Nr   r   rY   r   rZ   r  )r   rZ   r   r8   Úadd)Ú	enumerater  r`   ra   r   rb   r­   r5   r  r{   Ú	unsqueezeÚ	expand_asr  r&   re   rý   r
  )rR   r1   Úpatch_heightÚpatch_widthÚoutr  Úhidden_staterP   r~   Úsequence_lengthr@   r\   Úfeature_shapeÚreadouts                 r,   r…   zDPTReassembleStage.forward
  s  € ð ˆå(¨Ñ7Ô7ð 	%ñ 	%‰OˆAˆ|Ø˜Ô/Ð/Ñ/à*6°q°q°q¸!°tÔ*<¸lÈ1È1È1ÈaÈbÈbÈ5Ô>Q˜<�	Ø<HÔ<NÑ9�
˜O¨\ØÐ+°Ð0GØ#/×#7Ò#7¸
ÀLÐR]Ð_kÑ#lÔ#l�L�Lå$ _°cÑ%9Ñ:Ô:�DØ#/×#7Ò#7¸
ÀDÈ$ÐP\Ñ#]Ô#]�LØ+×3Ò3°A°q¸!¸QÑ?Ô?×JÒJÑLÔL�à ,Ô 2�Ø”;Ô+¨yÒ8Ð8à#/×#7Ò#7¸Ñ#:Ô#:×#BÒ#BÀ9Ñ#MÔ#M�LØ'×1Ò1°!Ñ4Ô4×>Ò>¸|ÑLÔL�Gà#; 4Ô#8¸Ô#;½E¼IÀ|ÐU\ÐF]Ð_aÑ<bÔ<bÑ#cÔ#c�Là#/×#7Ò#7¸¸1¸aÑ#@Ô#@×#HÒ#HÈÑ#WÔ#W�L�LØ”[Ô-°Ò6Ð6Ø#/×#7Ò#7¸Ñ#:Ô#:¸Y×=PÒ=PÐQSÑ=TÔ=TÑ#T�LØ#/×#7Ò#7¸Ñ#FÔ#F�LØ-˜tœ{¨1œ~¨lÑ;Ô;�Ø�JŠJ�|Ñ$Ô$Ð$Ñ$àˆ
r+   ©NN)r"   r#   r$   r%   r=   rÿ   r   Úlistr&   rŠ   r…   rŒ   r�   s   @r,   rú   rú   È  s˜   ø€ € € € € ðð ð
<ð 
<ð 
<ð 
<ð 
<ðð ð ð4
ð 
ð 
ð#ð # T¨%¬,Ô%7ð #ÐaeÐfkÔfrÔasð #ð #ð #ð #ð #ð #ð #ð #r+   rú   c                 ó`   — | j         �!t          | j         d¦  «        r| j         j        S | j        S )NrA   )Úbackbone_configr¸   rA   )r5   s    r,   r  r  0  s2   € ØÔÐ)­g°fÔ6LÈmÑ.\Ô.\Ð)ØÔ%Ô1Ð1àÔ!Ð!r+   c                   ó2   ‡ — e Zd Zdededefˆ fd„Zd„ Zˆ xZS )r  r5   rG   r  c           	      ó–  •— t          ¦   «                              ¦   «          t          |¦  «        }t          j        ||d¬¦  «        | _        |dk    r t          j        ||||d¬¦  «        | _        d S |dk    rt          j        ¦   «         | _        d S |dk     r0t          j        ||dt          d|z  ¦  «        d¬¦  «        | _        d S d S )Nr   )Úin_channelsÚout_channelsr;   r   ©r;   rœ   Úpaddingr   )
r<   r=   r  r   rL   rM   ÚConvTranspose2dÚresizer  r‰   )rR   r5   rG   r  rA   rV   s        €r,   r=   zDPTReassembleLayer.__init__8  sÅ   ø€ Ý‰Œ×ÒÑÔÐå/°Ñ7Ô7ˆÝœ)°È(Ð`aÐbÑbÔbˆŒð �AŠ:ˆ:ÝÔ,¨X°xÈVÐ\bÐlmÐnÑnÔnˆDŒKˆKˆKØ�qŠ[ˆ[Ýœ+™-œ-ˆDŒKˆKˆKØ�aŠZˆZåœ) H¨hÀAÍcÐRSÐV\ÑR\ÉoÌoÐghÐiÑiÔiˆDŒKˆKˆKð ˆZr+   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S r†   )rM   r*  )rR   r  s     r,   r…   zDPTReassembleLayer.forwardG  s*   € Ø—’ |Ñ4Ô4ˆØ—{’{ <Ñ0Ô0ˆØÐr+   )r"   r#   r$   r   r‰   r=   r…   rŒ   r�   s   @r,   r  r  7  sj   ø€ € € € € ðj˜yð j°Cð jÀð jð jð jð jð jð jðð ð ð ð ð ð r+   r  c                   ó*   ‡ — e Zd Zdefˆ fd„Zd„ Zˆ xZS )ÚDPTFeatureFusionStager5   c                 ó  •— t          ¦   «                              ¦   «          t          j        ¦   «         | _        t          t          |j        ¦  «        ¦  «        D ])}| j                             t          |¦  «        ¦  «         Œ*d S r†   )
r<   r=   r   rü   rý   r  rH   r  r
  ÚDPTFeatureFusionLayer)rR   r5   r™   rV   s      €r,   r=   zDPTFeatureFusionStage.__init__N  st   ø€ Ý‰Œ×ÒÑÔÐÝ”m‘o”oˆŒÝ•s˜6Ô3Ñ4Ô4Ñ5Ô5ð 	>ð 	>ˆAØŒK×ÒÕ4°VÑ<Ô<Ñ=Ô=Ð=Ð=ð	>ð 	>r+   c                 ó¸   — |d d d…         }g }d }t          || j        ¦  «        D ]4\  }}|€ ||¦  «        }n |||¦  «        }|                     |¦  «         Œ5|S )Nr8   )r  rý   r
  )rR   r1   Úfused_hidden_statesÚfused_hidden_stater  Úlayers         r,   r…   zDPTFeatureFusionStage.forwardT  sˆ   € à% d d¨ dÔ+ˆà ÐØ!ÐÝ#& }°d´kÑ#BÔ#Bð 	;ð 	;ÑˆL˜%Ø!Ð)à%* U¨<Ñ%8Ô%8Ð"Ð"à%* UÐ+=¸|Ñ%LÔ%LÐ"Ø×&Ò&Ð'9Ñ:Ô:Ð:Ð:à"Ð"r+   )r"   r#   r$   r   r=   r…   rŒ   r�   s   @r,   r-  r-  M  sS   ø€ € € € € ð>˜yð >ð >ð >ð >ð >ð >ð#ð #ð #ð #ð #ð #ð #r+   r-  c                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Zˆ xZ	S )ÚDPTPreActResidualLayerz©
    ResidualConvUnit, pre-activate residual unit.

    Args:
        config (`[DPTConfig]`):
            Model configuration class defining the model architecture.
    r5   c                 ó  •— t          ¦   «                              ¦   «          |j        | _        |j        �|j        n| j         }t          j        ¦   «         | _        t          j        |j	        |j	        ddd|¬¦  «        | _
        t          j        ¦   «         | _        t          j        |j	        |j	        ddd|¬¦  «        | _        | j        r>t          j        |j	        ¦  «        | _        t          j        |j	        ¦  «        | _        d S d S )Nr   r   )r;   rœ   r(  r¶   )r<   r=   Ú!use_batch_norm_in_fusion_residualÚuse_batch_normÚuse_bias_in_fusion_residualr   ÚReLUÚactivation1rL   Úfusion_hidden_sizeÚconvolution1Úactivation2Úconvolution2ÚBatchNorm2dÚbatch_norm1Úbatch_norm2)rR   r5   r9  rV   s      €r,   r=   zDPTPreActResidualLayer.__init__n  s  ø€ Ý‰Œ×ÒÑÔÐà$ÔFˆÔð Ô1Ð=ð Ô.Ð.àÔ(Ð(ð 	$õ œ7™9œ9ˆÔÝœIØÔ%ØÔ%ØØØØ,ð
ñ 
ô 
ˆÔõ œ7™9œ9ˆÔÝœIØÔ%ØÔ%ØØØØ,ð
ñ 
ô 
ˆÔð Ôð 	IÝ!œ~¨fÔ.GÑHÔHˆDÔÝ!œ~¨fÔ.GÑHÔHˆDÔÐÐð	Ið 	Ir+   r  rq   c                 ó(  — |}|                       |¦  «        }|                      |¦  «        }| j        r|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        r|                      |¦  «        }||z   S r†   )r;  r=  r8  rA  r>  r?  rB  ©rR   r  Úresiduals      r,   r…   zDPTPreActResidualLayer.forward�  sš   € ØˆØ×'Ò'¨Ñ5Ô5ˆà×(Ò(¨Ñ6Ô6ˆàÔð 	:Ø×+Ò+¨LÑ9Ô9ˆLà×'Ò'¨Ñ5Ô5ˆØ×(Ò(¨Ñ6Ô6ˆàÔð 	:Ø×+Ò+¨LÑ9Ô9ˆLà˜hÑ&Ð&r+   rž   r�   s   @r,   r5  r5  e  s|   ø€ € € € € ðð ð I˜yð  Ið  Ið  Ið  Ið  Ið  IðD' E¤Lð '°U´\ð 'ð 'ð 'ð 'ð 'ð 'ð 'ð 'r+   r5  c                   óh   ‡ — e Zd ZdZddedefˆ fd„Zddej        dej        dz  d	ej        fd
„Z	ˆ xZ
S )r/  a3  Feature fusion layer, merges feature maps from different stages.

    Args:
        config (`[DPTConfig]`):
            Model configuration class defining the model architecture.
        align_corners (`bool`, *optional*, defaults to `True`):
            The align_corner setting for bilinear upsample.
    Tr5   Úalign_cornersc                 óô   •— t          ¦   «                              ¦   «          || _        t          j        |j        |j        dd¬¦  «        | _        t          |¦  «        | _        t          |¦  «        | _	        d S )Nr   T)r;   r¶   )
r<   r=   rG  r   rL   r<  rM   r5  Úresidual_layer1Úresidual_layer2)rR   r5   rG  rV   s      €r,   r=   zDPTFeatureFusionLayer.__init__¬  si   ø€ Ý‰Œ×ÒÑÔÐà*ˆÔåœ) FÔ$=¸vÔ?XÐfgÐnrÐsÑsÔsˆŒå5°fÑ=Ô=ˆÔÝ5°fÑ=Ô=ˆÔÐÐr+   Nr  rE  rq   c                 ót  — |�c|j         |j         k    r;t          j                             ||j         d         |j         d         fdd¬¦  «        }||                      |¦  «        z   }|                      |¦  «        }t          j                             |dd| j        ¬¦  «        }|                      |¦  «        }|S )NrZ   r   r[   F©r\   r]   rG  ©Úscale_factorr]   rG  )r`   r   rc   rd   rI  rJ  rG  rM   rD  s      r,   r…   zDPTFeatureFusionLayer.forward¶  sÂ   € ØÐØÔ! X¤^Ò3Ð3Ýœ=×4Ò4Ø LÔ$6°qÔ$9¸<Ô;MÈaÔ;PÐ#QÐXbÐrwð 5ñ ô �ð (¨$×*>Ò*>¸xÑ*HÔ*HÑHˆLà×+Ò+¨LÑ9Ô9ˆÝ”}×0Ò0Ø q¨zÈÔI[ð 1ñ 
ô 
ˆð —’ |Ñ4Ô4ˆàÐr+   ©Tr†   )r"   r#   r$   r%   r   r‹   r=   r&   rŠ   r…   rŒ   r�   s   @r,   r/  r/  ¢  s•   ø€ € € € € ðð ð>ð >˜yð >¸ð >ð >ð >ð >ð >ð >ðð  E¤Lð ¸E¼LÈ4Ñ<Oð Ð[`Ô[gð ð ð ð ð ð ð ð r+   r/  c                   ó~   ‡ — e Zd ZU eed<   dZdZdZdZdZ	dZ
dZdZeedœZ ej        ¦   «         ˆ fd„¦   «         Zˆ xZS )ÚDPTPreTrainedModelr5   Údptro   )ÚimageT)r1   r2   c                 óê   •— t          ¦   «                              |¦  «         t          |t          t          f¦  «        r4t          j        |j        ¦  «         t          j        |j        ¦  «         dS dS )zInitialize the weightsN)	r<   Ú_init_weightsrB   r�   r4   ÚinitÚzeros_rP   rQ   )rR   r    rV   s     €r,   rU  z DPTPreTrainedModel._init_weights×  sl   ø€ õ 	‰Œ×Ò˜fÑ%Ô%Ð%Ý�fÕ/Õ1GÐHÑIÔIð 	4ÝŒK˜Ô(Ñ)Ô)Ð)ÝŒK˜Ô2Ñ3Ô3Ð3Ð3Ð3ð	4ð 	4r+   )r"   r#   r$   r   r(   Úbase_model_prefixÚmain_input_nameÚinput_modalitiesÚsupports_gradient_checkpointingÚ_supports_sdpaÚ_supports_flash_attnÚ_supports_flex_attnÚ_supports_attention_backendrê   r²   Ú_can_record_outputsr&   Úno_gradrU  rŒ   r�   s   @r,   rQ  rQ  Ç  s›   ø€ € € € € € àÐÐÑØÐØ$€OØ!ÐØ&*Ð#Ø€NØÐØÐØ"&Ðà$Ø&ðð Ðð
 €U„]�_„_ð4ð 4ð 4ð 4ñ „_ð4ð 4ð 4ð 4ð 4r+   rQ  c            	       óV   ‡ — e Zd Zdefˆ fd„Z	 d	dej        dedee	         de
fd„Zˆ xZS )
ÚDPTViTEncoderr5   c                 óÆ   •‡— t          ¦   «                              ¦   «          ‰| _        t          j        ˆfd„t          ‰j        ¦  «        D ¦   «         ¦  «        | _        d S )Nc                 ó.   •— g | ]}t          ‰¦  «        ‘ŒS r*   )rê   )rw   r™   r5   s     €r,   rz   z*DPTViTEncoder.__init__.<locals>.<listcomp>ä  s!   ø€ Ð#aÐ#aÐ#a¸A¥K°Ñ$7Ô$7Ð#aÐ#aÐ#ar+   )r<   r=   r5   r   rü   r  Únum_hidden_layersr3  rÀ   s    `€r,   r=   zDPTViTEncoder.__init__á  sV   øø€ Ý‰Œ×ÒÑÔÐØˆŒÝ”]Ð#aÐ#aÐ#aÐ#aÅÀvÔG_ÑA`ÔA`Ð#aÑ#aÔ#aÑbÔbˆŒ
ˆ
ˆ
r+   Fr1   r‚   r¦   rq   c                 óL   — | j         D ]} ||¦  «        }Œt          |¬¦  «        S )N)r/   )r3  r   )rR   r1   r‚   r¦   Úlayer_modules        r,   r…   zDPTViTEncoder.forwardæ  s7   € ð !œJð 	8ð 	8ˆLØ(˜L¨Ñ7Ô7ˆMˆMå°Ð?Ñ?Ô?Ð?r+   rˆ   )r"   r#   r$   r   r=   r&   rŠ   r‹   r   r   r   r…   rŒ   r�   s   @r,   rc  rc  à  s    ø€ € € € € ðc˜yð cð cð cð cð cð cð INð@ð @Ø"œ\ð@ØAEð@ØY_Ð`rÔYsð@à	ð@ð @ð @ð @ð @ð @ð @ð @r+   rc  c            	       óž   ‡ — e Zd Zddedefˆ fd„Zd„ Ze ed¬¦  «        e	de
j        d	ee         d
efd„¦   «         ¦   «         ¦   «         Zˆ xZS )ÚDPTModelTr5   Úadd_pooling_layerc                 ó‚  •— t          ¦   «                              |¦  «         || _        |j        rt	          |¦  «        | _        nt          |¦  «        | _        t          |¦  «        | _        t          j
        |j        |j        ¬¦  «        | _        |rt          |¦  «        nd| _        |                      ¦   «          dS )zv
        add_pooling_layer (bool, *optional*, defaults to `True`):
            Whether to add a pooling layer
        rì   N)r<   r=   r5   rþ   r4   rƒ   r�   rc  Úencoderr   rñ   rA   rò   Ú	layernormÚDPTViTPoolerÚpoolerÚ	post_init)rR   r5   rk  rV   s      €r,   r=   zDPTModel.__init__ñ  s­   ø€ õ
 	‰Œ×Ò˜Ñ Ô Ð ØˆŒð Ôð 	7Ý4°VÑ<Ô<ˆDŒOˆOå.¨vÑ6Ô6ˆDŒOÝ$ VÑ,Ô,ˆŒåœ fÔ&8¸fÔ>SÐTÑTÔTˆŒØ.?ÐI•l 6Ñ*Ô*Ð*ÀTˆŒð 	�ŠÑÔÐÐÐr+   c                 ó@   — | j         j        r| j        S | j        j        S r†   )r5   rþ   rƒ   r’   )rR   s    r,   Úget_input_embeddingszDPTModel.get_input_embeddings  s"   € ØŒ;Ô ð 	4Ø”?Ð"à”?Ô3Ð3r+   F)Útie_last_hidden_statesro   r¦   rq   c                 óø   — |                       |¦  «        }|j        } | j        |fi |¤Ž}|j        }|                      |¦  «        }| j        �|                      |¦  «        nd }t          |||j        ¬¦  «        S )N)r/   r0   r!   )rƒ   r    rm  r/   rn  rp  r.   r!   )rR   ro   r¦   Úembedding_outputÚembedding_last_hidden_statesÚencoder_outputsÚsequence_outputÚpooled_outputs           r,   r…   zDPTModel.forward  s—   € ð HLÇÂÐWcÑGdÔGdÐØ'7Ô'JÐ$à+7¨4¬<Ð8TÐ+_Ð+_ÐX^Ð+_Ð+_ˆØ)Ô;ˆàŸ.š.¨Ñ9Ô9ˆØ8<¼Ð8O˜Ÿš OÑ4Ô4Ð4ÐUYˆåCØ-Ø'Ø%5Ô%Nð
ñ 
ô 
ð 	
r+   rO  )r"   r#   r$   r   r‹   r=   rs  r   r   r   r&   r'   r   r   r.   r…   rŒ   r�   s   @r,   rj  rj  ï  sÇ   ø€ € € € € ðð ˜yð ¸Tð ð ð ð ð ð ð*4ð 4ð 4ð  Ø€_¨EÐ2Ñ2Ô2Øð
àÔ'ð
ð Ð+Ô,ð
ð 
>ð	
ð 
ð 
ñ „^ñ 3Ô2ñ  Ôð
ð 
ð 
ð 
ð 
r+   rj  c                   óH   ‡ — e Zd Zdefˆ fd„Zdej        dej        fd„Zˆ xZS )ro  r5   c                 ó¾   •— t          ¦   «                              ¦   «          t          j        |j        |j        ¦  «        | _        t          |j                 | _	        d S r†   )
r<   r=   r   r¾   rA   Úpooler_output_sizerÐ   r	   Ú
pooler_actÚ
activationrÀ   s     €r,   r=   zDPTViTPooler.__init__&  sE   ø€ Ý‰Œ×ÒÑÔÐÝ”Y˜vÔ1°6Ô3LÑMÔMˆŒ
Ý  Ô!2Ô3ˆŒˆˆr+   r1   rq   c                 ór   — |d d …df         }|                       |¦  «        }|                      |¦  «        }|S )Nr   )rÐ   r  )rR   r1   Úfirst_token_tensorrz  s       r,   r…   zDPTViTPooler.forward+  s@   € ð +¨1¨1¨1¨a¨4Ô0ÐØŸ
š
Ð#5Ñ6Ô6ˆØŸš¨Ñ6Ô6ˆØÐr+   rä   r�   s   @r,   ro  ro  %  sj   ø€ € € € € ð4˜yð 4ð 4ð 4ð 4ð 4ð 4ð
 U¤\ð °e´lð ð ð ð ð ð ð ð r+   ro  c            
       ó~   ‡ — e Zd ZdZdefˆ fd„Z	 	 d
deej                 de	dz  de	dz  deej                 fd	„Z
ˆ xZS )ÚDPTNecka;  
    DPTNeck. A neck is a module that is normally used between the backbone and the head. It takes a list of tensors as
    input and produces another list of tensors as output. For DPT, it includes 2 stages:

    * DPTReassembleStage
    * DPTFeatureFusionStage.

    Args:
        config (dict): config dict.
    r5   c           
      ó”  •— t          ¦   «                              ¦   «          || _        |j        �|j        j        dk    rd | _        nt          |¦  «        | _        t          j        ¦   «         | _	        |j
        D ]8}| j	                             t          j        ||j        ddd¬¦  «        ¦  «         Œ9t          |¦  «        | _        d S )NÚswinv2r   r   F©r;   r(  r¶   )r<   r=   r5   r"  Ú
model_typeÚreassemble_stagerú   r   rü   Úconvsr  r
  rL   r<  r-  Úfusion_stage)rR   r5   ÚchannelrV   s      €r,   r=   zDPTNeck.__init__@  sÄ   ø€ Ý‰Œ×ÒÑÔÐØˆŒð Ô!Ð-°&Ô2HÔ2SÐW_Ò2_Ð2_Ø$(ˆDÔ!Ð!å$6°vÑ$>Ô$>ˆDÔ!å”]‘_”_ˆŒ
ØÔ/ð 	sð 	sˆGØŒJ×Ò�bœi¨°Ô1JÐXYÐcdÐkpÐqÑqÔqÑrÔrÐrÐrõ 2°&Ñ9Ô9ˆÔÐÐr+   Nr1   r  r  rq   c                 ól  ‡ — t          |t          t          f¦  «        st          d¦  «        ‚t	          |¦  «        t	          ‰ j        j        ¦  «        k    rt          d¦  «        ‚‰ j        �‰                      |||¦  «        }ˆ fd„t          |¦  «        D ¦   «         }‰  
                    |¦  «        }|S )zñ
        Args:
            hidden_states (`list[torch.FloatTensor]`, each of shape `(batch_size, sequence_length, hidden_size)` or `(batch_size, hidden_size, height, width)`):
                List of hidden states from the backbone.
        z2hidden_states should be a tuple or list of tensorszOThe number of hidden states should be equal to the number of neck hidden sizes.Nc                 óB   •— g | ]\  }} ‰j         |         |¦  «        ‘ŒS r*   )r‰  )rw   r  ÚfeaturerR   s      €r,   rz   z#DPTNeck.forward.<locals>.<listcomp>f  s-   ø€ ÐVÐVÐV©z¨q°'�M�D”J˜q”M 'Ñ*Ô*ÐVÐVÐVr+   )rB   r)   r   Ú	TypeErrorrH   r5   r  rI   rˆ  r  rŠ  )rR   r1   r  r  r�   rÙ   s   `     r,   r…   zDPTNeck.forwardQ  s·   ø€ õ ˜-­%µ¨Ñ7Ô7ð 	RÝÐPÑQÔQÐQåˆ}ÑÔ¥ T¤[Ô%BÑ!CÔ!CÒCÐCÝÐnÑoÔoÐoð Ô Ð,Ø ×1Ò1°-ÀÈ{Ñ[Ô[ˆMàVÐVÐVÐV½YÀ}Ñ=UÔ=UÐVÑVÔVˆð ×"Ò" 8Ñ,Ô,ˆàˆr+   r  )r"   r#   r$   r%   r   r=   r   r&   rŠ   r‰   r…   rŒ   r�   s   @r,   rƒ  rƒ  4  s«   ø€ € € € € ð	ð 	ð:˜yð :ð :ð :ð :ð :ð :ð( $(Ø"&ð	ð à˜EœLÔ)ðð ˜D‘jðð ˜4‘Zð	ð
 
ˆeŒlÔ	ðð ð ð ð ð ð ð r+   rƒ  c                   óX   ‡ — e Zd ZdZdefˆ fd„Zdeej                 dej        fd„Z	ˆ xZ
S )ÚDPTDepthEstimationHeada	  
    Output head consisting of 3 convolutional layers. It progressively halves the feature dimension and upsamples
    the predictions to the input resolution after the first convolutional layer (details can be found in the paper's
    supplementary material).
    r5   c                 óü  •— t          ¦   «                              ¦   «          || _        d | _        |j        rt          j        ddddd¬¦  «        | _        |j        }t          j        t          j        ||dz  ddd¬¦  «        t          j	        ddd	¬
¦  «        t          j        |dz  dddd¬¦  «        t          j
        ¦   «         t          j        ddddd¬¦  «        t          j
        ¦   «         ¦  «        | _        d S )Né   )r   r   )r   r   r'  rZ   r   r   r[   TrM  é    r   )r<   r=   r5   rM   Úadd_projectionr   rL   r<  r  ÚUpsampler:  Úhead©rR   r5   r�   rV   s      €r,   r=   zDPTDepthEstimationHead.__init__u  sæ   ø€ Ý‰Œ×ÒÑÔÐàˆŒàˆŒØÔ ð 	eÝ œi¨¨S¸fÈVÐ]cÐdÑdÔdˆDŒOàÔ,ˆÝ”MÝŒI�h ¨A¡¸1ÀQÐPQÐRÑRÔRÝŒK Q¨ZÀtÐLÑLÔLÝŒI�h !‘m R°Q¸qÈ!ÐLÑLÔLÝŒG‰IŒIÝŒI�b˜!¨°1¸aÐ@Ñ@Ô@ÝŒG‰IŒIñ
ô 
ˆŒ	ˆ	ˆ	r+   r1   rq   c                 óð   — || j         j                 }| j        �1|                      |¦  «        } t          j        ¦   «         |¦  «        }|                      |¦  «        }|                     d¬¦  «        }|S )Nr   r^   )r5   Úhead_in_indexrM   r   r:  r—  Úsqueeze)rR   r1   Úpredicted_depths      r,   r…   zDPTDepthEstimationHead.forwardˆ  sl   € à% d¤kÔ&?Ô@ˆàŒ?Ð&Ø ŸOšO¨MÑ:Ô:ˆMØ%�BœG™IœI mÑ4Ô4ˆMàŸ)š) MÑ2Ô2ˆØ)×1Ò1°aÐ1Ñ8Ô8ˆàÐr+   )r"   r#   r$   r%   r   r=   r   r&   rŠ   r…   rŒ   r�   s   @r,   r‘  r‘  n  sy   ø€ € € € € ðð ð
˜yð 
ð 
ð 
ð 
ð 
ð 
ð& T¨%¬,Ô%7ð ¸E¼Lð ð ð ð ð ð ð ð r+   r‘  zu
    DPT Model with a depth estimation head on top (consisting of 3 convolutional layers) e.g. for KITTI, NYUv2.
    c                   ó�   ‡ — e Zd Zˆ fd„Zeee	 ddej        dej	        dz  de
e         defd„¦   «         ¦   «         ¦   «         Zˆ xZS )	ÚDPTForDepthEstimationc                 óF  •— t          ¦   «                              |¦  «         d | _        |j        du r|j        �t          |¦  «        | _        nt          |d¬¦  «        | _        t          |¦  «        | _	        t          |¦  «        | _        |                      ¦   «          d S ©NF)rk  )r<   r=   rF   rþ   r"  r   rj  rR  rƒ  Úneckr‘  r—  rq  rÀ   s     €r,   r=   zDPTForDepthEstimation.__init__œ  s•   ø€ Ý‰Œ×Ò˜Ñ Ô Ð àˆŒØÔ˜uÐ$Ð$¨Ô)?Ð)KÝ)¨&Ñ1Ô1ˆDŒMˆMå ¸%Ð@Ñ@Ô@ˆDŒHõ ˜F‘O”OˆŒ	õ +¨6Ñ2Ô2ˆŒ	ð 	�ŠÑÔÐÐÐr+   Nro   Úlabelsr¦   rq   c                 ó”  ‡ — d}|�t          d¦  «        ‚‰ j        � ‰ j        j        |fi |¤Ž}|j        }n„ ‰ j        |fi |¤Ž}|j        }‰ j        j        s$ˆ fd„t          |dd…         ¦  «        D ¦   «         }n?|j	        }| 
                    ˆ fd„t          |dd…         ¦  «        D ¦   «         ¦  «         |}d\  }}	‰ j        j        �5‰ j        j        du r'|j        \  }
}
}}‰ j        j        j        }||z  }||z  }	‰                      |||	¦  «        }‰                      |¦  «        }t!          |||j        |j        ¬¦  «        S )	aÙ  
        labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*):
            Ground truth depth estimation maps for computing the loss.

        Examples:
        ```python
        >>> from transformers import AutoImageProcessor, DPTForDepthEstimation
        >>> import torch
        >>> import numpy as np
        >>> 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("Intel/dpt-large")
        >>> model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")

        >>> # prepare image for the model
        >>> inputs = image_processor(images=image, return_tensors="pt")

        >>> with torch.no_grad():
        ...     outputs = model(**inputs)

        >>> # interpolate to original size
        >>> post_processed_output = image_processor.post_process_depth_estimation(
        ...     outputs,
        ...     target_sizes=[(image.height, image.width)],
        ... )

        >>> # visualize the prediction
        >>> predicted_depth = post_processed_output[0]["predicted_depth"]
        >>> depth = predicted_depth * 255 / predicted_depth.max()
        >>> depth = depth.detach().cpu().numpy()
        >>> depth = Image.fromarray(depth.astype("uint8"))
        ```NzTraining is not implemented yetc                 ó6   •— g | ]\  }}|‰j         j        v ¯|‘ŒS r*   ©r5   Úbackbone_out_indices©rw   ÚidxrŽ  rR   s      €r,   rz   z1DPTForDepthEstimation.forward.<locals>.<listcomp>ê  s6   ø€ ð !ð !ð !Ù ,  WÐPSÐW[ÔWbÔWwÐPwÐPw�GÐPwÐPwÐPwr+   r   c              3   óN   •K  — | ]\  }}|‰j         j        d d…         v ¯|V — Œ dS ©rZ   Nr¥  r§  s      €r,   ú	<genexpr>z0DPTForDepthEstimation.forward.<locals>.<genexpr>ï  sL   øè è € ð .ð .á$˜˜WØ˜dœkÔ>¸q¸r¸rÔBÐBÐBð àBÐBÐBÐBð.ð .r+   r  F)Úlossrœ  r1   r2   )ÚNotImplementedErrorrF   Úforward_with_filtered_kwargsrv   rR  r1   r5   rþ   r  r!   Úextendr"  r`   r?   r¡  r—  r   r2   )rR   ro   r¢  r¦   r¬  Úoutputsr1   Úbackbone_hidden_statesr  r  r™   r   r€   r?   rœ  s   `              r,   r…   zDPTForDepthEstimation.forward®  sÁ  ø€ ð^ ˆØÐÝ%Ð&GÑHÔHÐHàŒ=Ð$Ø@�d”mÔ@ÀÐXÐXÐQWÐXÐXˆGØ#Ô0ˆMˆMà�d”h˜|Ð6Ð6¨vÐ6Ð6ˆGØ#Ô1ˆMð ”;Ô(ð 7ð!ð !ð !ð !Ý09¸-ÈÈÈÔ:KÑ0LÔ0Lð!ñ !ô !��ð *1Ô)IÐ&Ø&×-Ò-ð .ð .ð .ð .å(1°-ÀÀÀÔ2CÑ(DÔ(Dð.ñ .ô .ñ ô ð ð
 !7�à$.Ñ!ˆ�kØŒ;Ô&Ð2°t´{Ô7LÐPUÐ7UÐ7UØ".Ô"4ÑˆAˆq�&˜%ØœÔ4Ô?ˆJØ! ZÑ/ˆLØ :Ñ-ˆKàŸ	š	 -°¸{ÑKÔKˆØŸ)š) MÑ2Ô2ˆå#ØØ+Ø!Ô/ØÔ)ð	
ñ 
ô 
ð 	
r+   r†   )r"   r#   r$   r=   r   r   r
   r&   r'   Ú
LongTensorr   r   r   r…   rŒ   r�   s   @r,   rž  rž  –  s¼   ø€ € € € € ðð ð ð ð ð$ ØØ ð +/ðT
ð T
àÔ'ðT
ð Ô  4Ñ'ðT
ð Ð+Ô,ð	T
ð
 
ðT
ð T
ð T
ñ !Ô ñ „^ñ ÔðT
ð T
ð T
ð T
ð T
r+   rž  c                   óT   ‡ — e Zd Zdefˆ fd„Zdeej                 dej        fd„Zˆ xZ	S )ÚDPTSemanticSegmentationHeadr5   c                 ó   •— t          ¦   «                              ¦   «          || _        |j        }t	          j        t	          j        ||ddd¬¦  «        t	          j        |¦  «        t	          j        ¦   «         t	          j	        |j
        ¦  «        t	          j        ||j        d¬¦  «        t	          j        ddd¬	¦  «        ¦  «        | _        d S )
Nr   r   Fr†  r:   rZ   r[   TrM  )r<   r=   r5   r<  r   r  rL   r@  r:  r“   Úsemantic_classifier_dropoutÚ
num_labelsr–  r—  r˜  s      €r,   r=   z$DPTSemanticSegmentationHead.__init__	  s¨   ø€ Ý‰Œ×ÒÑÔÐàˆŒØÔ,ˆÝ”MÝŒI�h °aÀÈÐOÑOÔOÝŒN˜8Ñ$Ô$ÝŒG‰IŒIÝŒJ�vÔ9Ñ:Ô:ÝŒI�h Ô 1¸qÐAÑAÔAÝŒK Q¨ZÀtÐLÑLÔLñ
ô 
ˆŒ	ˆ	ˆ	r+   r1   rq   c                 óT   — || j         j                 }|                      |¦  «        }|S r†   )r5   rš  r—  ©rR   r1   Úlogitss      r,   r…   z#DPTSemanticSegmentationHead.forward  s'   € à% d¤kÔ&?Ô@ˆØ—’˜=Ñ)Ô)ˆØˆr+   )
r"   r#   r$   r   r=   r   r&   rŠ   r…   rŒ   r�   s   @r,   r´  r´    so   ø€ € € € € ð
˜yð 
ð 
ð 
ð 
ð 
ð 
ð T¨%¬,Ô%7ð ¸E¼Lð ð ð ð ð ð ð ð r+   r´  c                   óH   ‡ — e Zd Zdefˆ fd„Zdej        dej        fd„Zˆ xZS )ÚDPTAuxiliaryHeadr5   c                 ó^  •— t          ¦   «                              ¦   «          |j        }t          j        t          j        ||ddd¬¦  «        t          j        |¦  «        t          j        ¦   «         t          j        dd¦  «        t          j        ||j	        d¬¦  «        ¦  «        | _
        d S )Nr   r   Fr†  gš™™™™™¹?r:   )r<   r=   r<  r   r  rL   r@  r:  r“   r·  r—  r˜  s      €r,   r=   zDPTAuxiliaryHead.__init__  sŒ   ø€ Ý‰Œ×ÒÑÔÐàÔ,ˆÝ”MÝŒI�h °aÀÈÐOÑOÔOÝŒN˜8Ñ$Ô$ÝŒG‰IŒIÝŒJ�s˜EÑ"Ô"ÝŒI�h Ô 1¸qÐAÑAÔAñ
ô 
ˆŒ	ˆ	ˆ	r+   r1   rq   c                 ó0   — |                       |¦  «        }|S r†   )r—  r¹  s      r,   r…   zDPTAuxiliaryHead.forward+  s   € Ø—’˜=Ñ)Ô)ˆØˆr+   rä   r�   s   @r,   r¼  r¼    sj   ø€ € € € € ð

˜yð 

ð 

ð 

ð 

ð 

ð 

ð U¤\ð °e´lð ð ð ð ð ð ð ð r+   r¼  c                   óž   ‡ — e Zd Zdefˆ fd„Zeee	 	 d	dej	        dz  dej
        dz  dee         defd„¦   «         ¦   «         ¦   «         Zˆ xZS )
ÚDPTForSemanticSegmentationr5   c                 ó(  •— t          ¦   «                              |¦  «         t          |d¬¦  «        | _        t	          |¦  «        | _        t          |¦  «        | _        |j        rt          |¦  «        nd | _
        |                      ¦   «          d S r   )r<   r=   rj  rR  rƒ  r¡  r´  r—  Úuse_auxiliary_headr¼  Úauxiliary_headrq  rÀ   s     €r,   r=   z#DPTForSemanticSegmentation.__init__2  s†   ø€ Ý‰Œ×Ò˜Ñ Ô Ð å˜F°eÐ<Ñ<Ô<ˆŒõ ˜F‘O”OˆŒ	õ 0°Ñ7Ô7ˆŒ	Ø:@Ô:SÐ]Õ.¨vÑ6Ô6Ð6ÐY]ˆÔð 	�ŠÑÔÐÐÐr+   Nro   r¢  r¦   rq   c                 óz  ‡ — |�‰ j         j        dk    rt          d¦  «        ‚ ‰ j        |fi |¤Ž}|j        }‰ j         j        s$ˆ fd„t          |dd…         ¦  «        D ¦   «         }n?|j        }|                     ˆ fd„t          |dd…         ¦  «        D ¦   «         ¦  «         |}‰  	                    |¬¦  «        }‰  
                    |¦  «        }d}‰ j        �‰                      |d         ¦  «        }d}	|�¦t          j                             ||j        dd…         d	d
¬¦  «        }
|�0t          j                             ||j        dd…         d	d
¬¦  «        }t!          ‰ j         j        ¬¦  «        } ||
|¦  «        } |||¦  «        }|‰ j         j        |z  z   }	t'          |	||j        |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, DPTForSemanticSegmentation
        >>> 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("Intel/dpt-large-ade")
        >>> model = DPTForSemanticSegmentation.from_pretrained("Intel/dpt-large-ade")

        >>> inputs = image_processor(images=image, return_tensors="pt")

        >>> outputs = model(**inputs)
        >>> logits = outputs.logits
        ```Nr   z/The number of labels should be greater than onec                 ó6   •— g | ]\  }}|‰j         j        v ¯|‘ŒS r*   r¥  r§  s      €r,   rz   z6DPTForSemanticSegmentation.forward.<locals>.<listcomp>k  s5   ø€ ð ð ð Ù(˜C ÈCÐSWÔS^ÔSsÐLsÐLs�ÐLsÐLsÐLsr+   c              3   óN   •K  — | ]\  }}|‰j         j        d d…         v ¯|V — Œ dS rª  r¥  r§  s      €r,   r«  z5DPTForSemanticSegmentation.forward.<locals>.<genexpr>p  sM   øè è € ð *ð *Ù(˜C ÈCÐSWÔS^ÔSsÐtuÐtvÐtvÔSwÐLwÐLw�ÐLwÐLwÐLwÐLwð*ð *r+   )r1   r8   r9   r[   FrL  )Úignore_index)r¬  rº  r1   r2   )r5   r·  rI   rR  r1   rþ   r  r!   r¯  r¡  r—  rÃ  r   rc   rd   r`   r   Úsemantic_loss_ignore_indexÚauxiliary_loss_weightr   r2   )rR   ro   r¢  r¦   r°  r1   r±  rº  Úauxiliary_logitsr¬  Úupsampled_logitsÚupsampled_auxiliary_logitsÚloss_fctÚ	main_lossÚauxiliary_losss   `              r,   r…   z"DPTForSemanticSegmentation.forwardA  s3  ø€ ðB Ð $¤+Ô"8¸AÒ"=Ð"=ÝÐNÑOÔOÐOàHPÈÌÐQ]ÐHhÐHhÐagÐHhÐHhˆØÔ-ˆð Œ{Ô$ð 
	3ðð ð ð Ý,5°mÀAÀBÀBÔ6GÑ,HÔ,Hðñ ô ˆMˆMð &-Ô%EÐ"Ø"×)Ò)ð *ð *ð *ð *Ý,5°mÀAÀBÀBÔ6GÑ,HÔ,Hð*ñ *ô *ñ ô ð ð 3ˆMàŸ	š	°˜	Ñ>Ô>ˆØ—’˜=Ñ)Ô)ˆàÐØÔÐ*Ø#×2Ò2°=ÀÔ3DÑEÔEÐàˆØÐå!œ}×8Ò8Ø˜Vœ\¨"¨#¨#Ô.°ZÈuð  9ñ  ô  Ðð  Ð+Ý-/¬]×-FÒ-FØ$¨6¬<¸¸¸Ô+<À:Ð]bð .Gñ .ô .Ð*õ (°T´[Ô5[Ð\Ñ\Ô\ˆHØ ˜Ð!1°6Ñ:Ô:ˆIØ%˜XÐ&@À&ÑIÔIˆNØ˜tœ{Ô@À>ÑQÑQˆDå&ØØØ!Ô/ØÔ)ð	
ñ 
ô 
ð 	
r+   r  )r"   r#   r$   r   r=   r   r   r
   r&   r'   r²  r   r   r   r…   rŒ   r�   s   @r,   rÀ  rÀ  0  sÐ   ø€ € € € € ð˜yð ð ð ð ð ð ð ØØ ð 26Ø*.ðN
ð N
àÔ'¨$Ñ.ðN
ð Ô  4Ñ'ðN
ð Ð+Ô,ð	N
ð
 
!ðN
ð N
ð N
ñ !Ô ñ „^ñ ÔðN
ð N
ð N
ð N
ð N
r+   rÀ  )rž  rÀ  rj  rQ  )NrŸ   )Mr%   Úcollections.abcrC   r   Údataclassesr   r&   r   Útorch.nnr   Ú r   rV  Úactivationsr	   Úbackbone_utilsr
   r   Úmodeling_layersr   Úmodeling_outputsr   r   r   Úmodeling_utilsr   r   Úprocessing_utilsr   Úutilsr   r   r   r   r   Úutils.genericr   r   Úutils.output_capturingr   Úconfiguration_dptr   Ú
get_loggerr"   Úloggerr   r.   ÚModuler4   r�   r‘   rŠ   Úfloatr°   r²   rÎ   rÖ   rÝ   ræ   rê   rú   r  r  r-  r5  r/  rQ  rc  rj  ro  rƒ  r‘  rž  r´  r¼  rÀ  Ú__all__r*   r+   r,   ú<module>rã     s‹  ððð ð Ð Ð Ð Ø $Ð $Ð $Ð $Ð $Ð $Ø !Ð !Ð !Ð !Ð !Ð !à €€€Ø Ð Ð Ð Ð Ð Ø %Ð %Ð %Ð %Ð %Ð %à &Ð &Ð &Ð &Ð &Ð &Ø !Ð !Ð !Ð !Ð !Ð !Ø HÐ HÐ HÐ HÐ HÐ HÐ HÐ HØ 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø ^Ð ^Ð ^Ð ^Ð ^Ð ^Ð ^Ð ^Ð ^Ð ^Ø FÐ FÐ FÐ FÐ FÐ FÐ FÐ FØ &Ð &Ð &Ð &Ð &Ð &Ø XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XØ IÐ IÐ IÐ IÐ IÐ IÐ IÐ IØ 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø (Ð (Ð (Ð (Ð (Ð (ð 
ˆÔ	˜HÑ	%Ô	%€ð €ððñ ô ð ð	Jð 	Jð 	Jð 	Jð 	J°ñ 	Jô 	Jñ „ñô ð	Jð €ððñ ô ð ðJð Jð Jð Jð J¸;ñ Jô Jñ „ñô ðJð$]
ð ]
ð ]
ð ]
ð ]
˜RœYñ ]
ô ]
ð ]
ð@4Yð 4Yð 4Yð 4Yð 4Y�r”yñ 4Yô 4Yð 4Yðnð ð ð ð ˜BœIñ ô ð ðL !Øð%ð %ØŒIð%àŒ<ð%ð 
Œð%ð Œ<ð	%ð
 ”L 4Ñ'ð%ð �T‰\ð%ð ð%ð Ð'Ô(ð%ð %ð %ð %ð:4.ð 4.ð 4.ð 4.ð 4.�r”yñ 4.ô 4.ð 4.ðpð ð ð ð �r”yñ ô ð ð$ð ð ð ð �b”iñ ô ð ð"ð ð ð ð ˜œñ ô ð ð 
ð 
ð 
ð 
ð 
�2”9ñ 
ô 
ð 
ðð ð ð ð Ð,ñ ô ð ðDeð eð eð eð e˜œñ eô eð eðP"ð "ð "ðð ð ð ð ˜œñ ô ð ð,#ð #ð #ð #ð #˜BœIñ #ô #ð #ð0:'ð :'ð :'ð :'ð :'˜RœYñ :'ô :'ð :'ðz"ð "ð "ð "ð "˜BœIñ "ô "ð "ðJ ð4ð 4ð 4ð 4ð 4˜ñ 4ô 4ñ „ð4ð0@ð @ð @ð @ð @�B”Iñ @ô @ð @ð ð1
ð 1
ð 1
ð 1
ð 1
Ð!ñ 1
ô 1
ñ „ð1
ðjð ð ð ð �2”9ñ ô ð ð7ð 7ð 7ð 7ð 7ˆbŒiñ 7ô 7ð 7ðt%ð %ð %ð %ð %˜RœYñ %ô %ð %ðP €ððñ ô ð
j
ð j
ð j
ð j
ð j
Ð.ñ j
ô j
ñô ð
j
ðZð ð ð ð  "¤)ñ ô ð ð,ð ð ð ð �r”yñ ô ð ð$ ða
ð a
ð a
ð a
ð a
Ð!3ñ a
ô a
ñ „ða
ðH dÐ
cÐ
c€€€r+   