§
    ·¯¸j‚À  ã                   ód  — d Z ddlZddlmZmZmZmZmZmZm	Z	m
Z
 ddlZddlmZ ddlmc mZ ddlmZmZ ddlmZmZmZmZmZmZ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gZ% G d„ dej&        ¦  «        Z' G d„ dej&        ¦  «        Z( G d„ dej&        ¦  «        Z) G d„ dej&        ¦  «        Z* G d„ dej&        ¦  «        Z+ G d„ dej&        ¦  «        Z,de-dedej&        fd„Z.d_dee/df         de0de/dee/e/e/e/f         fd „Z1 G d!„ d"ej&        ¦  «        Z2 G d#„ d$ej&        ¦  «        Z3	 	 	 	 	 	 	 	 	 	 d`d*ed+e/d,e/d-ee/         d.e/d/e/d0e/d1e/d2e0d3e4d4e0d5e0dedej5        fd6„Z6	 	 	 	 dad*ed+e/d,e/d-ee/         d.e/d2e0d3e4d4e0d5e0dedej5        fd7„Z7 G d8„ dej&        ¦  «        Z8dbd9e-d:e4dede8fd;„Z9dcd=e-dedee-ef         fd>„Z: e$ e:d?d@dA¬B¦  «         e:d?dCdDdE¬F¦  «         e:d?dGdA¬B¦  «         e:d?dHdDdE¬F¦  «         e:d?dIdA¬B¦  «         e:d?dJdDdK¬F¦  «         e:d?dLdA¬B¦  «         e:d?dMdNdK¬F¦  «         e:d?dOdA¬B¦  «         e:d?dPdNdK¬F¦  «         e:d?dQdNdR¬F¦  «        dSœ¦  «        Z;e#dbd:e4dede8fdT„¦   «         Z<e#dbd:e4dede8fdU„¦   «         Z=e#dbd:e4dede8fdV„¦   «         Z>e#dbd:e4dede8fdW„¦   «         Z?e#dbd:e4dede8fdX„¦   «         Z@e#dbd:e4dede8fdY„¦   «         ZAe#dbd:e4dede8fdZ„¦   «         ZBe#dbd:e4dede8fd[„¦   «         ZCe#dbd:e4dede8fd\„¦   «         ZDe#dbd:e4dede8fd]„¦   «         ZEe#dbd:e4dede8fd^„¦   «         ZFdS )da5   Vision OutLOoker (VOLO) implementation

Paper: `VOLO: Vision Outlooker for Visual Recognition` - https://arxiv.org/abs/2106.13112

Code adapted from official impl at https://github.com/sail-sg/volo, original copyright in comment below

Modifications and additions for timm by / Copyright 2022, Ross Wightman
é    N)ÚAnyÚCallableÚDictÚListÚOptionalÚTupleÚUnionÚType©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚDropPathÚMlpÚ	to_2tupleÚ	to_ntupleÚtrunc_normal_Úuse_fused_attnÚget_device_dtypeé   )Úbuild_model_with_cfg)Úfeature_take_indices)Ú
checkpoint)Úregister_modelÚgenerate_default_cfgsÚVOLOc                   óz   ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 ddeded	ed
ededededefˆ fd„Zdej	        dej	        fd„Z
ˆ xZS )ÚOutlookAttentionz,Outlook attention mechanism for VOLO models.é   r   Fç        NÚdimÚ	num_headsÚkernel_sizeÚpaddingÚstrideÚqkv_biasÚ	attn_dropÚ	proj_dropc                 ó  •— |	|
dœ}t          ¦   «                              ¦   «          ||z  }|| _        || _        || _        || _        |dz  | _        t          j        ||fd|i|¤Ž| _	        t          j        ||dz  |z  fi |¤Ž| _
        t          j        |¦  «        | _        t          j        ||fi |¤Ž| _        t          j        |¦  «        | _        t          j        |||¬¦  «        | _        t          j        ||d¬¦  «        | _        dS )	aÙ  Initialize OutlookAttention.

        Args:
            dim: Input feature dimension.
            num_heads: Number of attention heads.
            kernel_size: Kernel size for attention computation.
            padding: Padding for attention computation.
            stride: Stride for attention computation.
            qkv_bias: Whether to use bias in linear layers.
            attn_drop: Attention dropout rate.
            proj_drop: Projection dropout rate.
        ©ÚdeviceÚdtypeç      à¿Úbiasé   )r"   r#   r$   T)r"   r$   Ú	ceil_modeN)ÚsuperÚ__init__r!   r"   r#   r$   ÚscaleÚnnÚLinearÚvÚattnÚDropoutr&   Úprojr'   ÚUnfoldÚunfoldÚ	AvgPool2dÚpool)Úselfr    r!   r"   r#   r$   r%   r&   r'   r*   r+   ÚddÚhead_dimÚ	__class__s                €ú`/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/volo.pyr1   zOutlookAttention.__init__*   s  ø€ ð2 ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ˜)Ñ#ˆØ"ˆŒØ&ˆÔØˆŒØˆŒØ Ñ%ˆŒ
å”˜3 Ð9Ð9¨(Ð9°bÐ9Ð9ˆŒÝ”I˜c ;°!Ñ#3°iÑ#?ÐFÐFÀ2ÐFÐFˆŒ	åœ IÑ.Ô.ˆŒÝ”I˜c 3Ð-Ð-¨"Ð-Ð-ˆŒ	Ýœ IÑ.Ô.ˆŒå”i¨KÀÐQWÐXÑXÔXˆŒÝ”L¨V¸FÈdÐSÑSÔSˆŒ	ˆ	ˆ	ó    ÚxÚreturnc                 ó¶  — |j         \  }}}}|                      |¦  «                             dddd¦  «        }t          j        || j        z  ¦  «        t          j        || j        z  ¦  «        }}|                      |¦  «                             || j        || j        z  | j	        | j	        z  ||z  ¦  «                             ddddd¦  «        }|  
                    |                     dddd¦  «        ¦  «                             dddd¦  «        }	|                      |	¦  «                             |||z  | j        | j	        | j	        z  | j	        | j	        z  ¦  «                             ddddd¦  «        }	|	| j        z  }	|	                     d¬¦  «        }	|                      |	¦  «        }	|	|z                       ddddd¦  «                             ||| j	        z  | j	        z  ||z  ¦  «        }t          j        |||f| j	        | j        | j        ¬¦  «        }|                      |                     dddd¦  «        ¦  «        }|                      |¦  «        }|S )	ú›Forward pass.

        Args:
            x: Input tensor of shape (B, H, W, C).

        Returns:
            Output tensor of shape (B, H, W, C).
        r   r   r   é   r.   éÿÿÿÿ©r    )Úoutput_sizer"   r#   r$   )Úshaper5   ÚpermuteÚmathÚceilr$   r:   Úreshaper!   r"   r<   r6   r2   Úsoftmaxr&   ÚFÚfoldr#   r8   r'   )
r=   rC   ÚBÚHÚWÚCr5   ÚhÚwr6   s
             rA   ÚforwardzOutlookAttention.forwardV   s/  € ð ”W‰
ˆˆ1ˆa�à�FŠF�1‰IŒI×Ò˜a  A qÑ)Ô)ˆåŒy˜˜Tœ[™Ñ)Ô)­4¬9°Q¸¼±_Ñ+EÔ+Eˆ1ˆØ�KŠK˜‰NŒN×"Ò"ØˆtŒ~˜q D¤NÑ2ØÔ˜tÔ/Ñ/°°Q±ñ8ô 8ç8?ºÀÀ1ÀaÈÈAÑ8NÔ8Nð 	
ð �yŠy˜Ÿš 1 a¨¨AÑ.Ô.Ñ/Ô/×7Ò7¸¸1¸aÀÑCÔCˆØ�yŠy˜‰Œ×&Ò&Øˆq�1‰u�d”n dÔ&6¸Ô9IÑ&IØÔ˜tÔ/Ñ/ñ1ô 1ç18²¸¸A¸qÀ!ÀQÑ1GÔ1Gð 	ð �d”jÑ ˆØ�|Š| ˆ|Ñ#Ô#ˆØ�~Š~˜dÑ#Ô#ˆà�A‰X×Ò˜q ! Q¨¨1Ñ-Ô-×5Ò5°a¸¸TÔ=MÑ9MÐPTÔP`Ñ9`ÐbcÐfgÑbgÑhÔhˆÝŒF�1 1 a &°dÔ6FÐPTÔP\ÐeiÔepÐqÑqÔqˆà�IŠI�a—i’i  1 a¨Ñ+Ô+Ñ,Ô,ˆØ�NŠN˜1ÑÔˆàˆrB   )r   r   r   Fr   r   NN)Ú__name__Ú
__module__Ú__qualname__Ú__doc__ÚintÚboolÚfloatr1   ÚtorchÚTensorrY   Ú__classcell__©r@   s   @rA   r   r   '   sæ   ø€ € € € € Ø6Ð6ð  !ØØØ"Ø!Ø!ØØð*Tð *Tàð*Tð ð*Tð ð	*Tð
 ð*Tð ð*Tð ð*Tð ð*Tð ð*Tð *Tð *Tð *Tð *Tð *TðX ˜œð  ¨%¬,ð  ð  ð  ð  ð  ð  ð  ð  rB   r   c                   óÊ   ‡ — e Zd ZdZdddddej        ej        dddf
deded	ed
ededededede	ej
                 de	ej
                 defˆ fd„Zdej        dej        fd„Zˆ xZS )Ú	Outlookerz9Outlooker block that combines outlook attention with MLP.r   ç      @r   FNr    r"   r#   r$   r!   Ú	mlp_ratior&   Ú	drop_pathÚ	act_layerÚ
norm_layerr%   c           	      óº  •— ||dœ}t          ¦   «                              ¦   «           |
|fi |¤Ž| _        t          ||f|||||dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _         |
|fi |¤Ž| _	        t          d|t          ||z  ¦  «        |	dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _        dS )af  Initialize Outlooker block.

        Args:
            dim: Input feature dimension.
            kernel_size: Kernel size for outlook attention.
            padding: Padding for outlook attention.
            stride: Stride for outlook attention.
            num_heads: Number of attention heads.
            mlp_ratio: Ratio for MLP hidden dimension.
            attn_drop: Attention dropout rate.
            drop_path: Stochastic depth drop rate.
            act_layer: Activation layer type.
            norm_layer: Normalization layer type.
            qkv_bias: Whether to use bias in linear layers.
        r)   )r"   r#   r$   r%   r&   r   ©Úin_featuresÚhidden_featuresrj   N© )r0   r1   Únorm1r   r6   r   r3   ÚIdentityÚ
drop_path1Únorm2r   r^   ÚmlpÚ
drop_path2)r=   r    r"   r#   r$   r!   rh   r&   ri   rj   rk   r%   r*   r+   r>   r@   s                  €rA   r1   zOutlooker.__init__|   s  ø€ ð> ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ�Z Ð*Ð* rÐ*Ð*ˆŒ
Ý$ØØð	
ð $ØØØØð	
ð 	
ð ð	
ð 	
ˆŒ	ð 2;¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒà�Z Ð*Ð* rÐ*Ð*ˆŒ
Ýð 
ØÝ  i¡Ñ0Ô0Øð
ð 
ð ð	
ð 
ˆŒð 2;¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒˆˆrB   rC   rD   c                 óþ   — ||                       |                      |                      |¦  «        ¦  «        ¦  «        z   }||                      |                      |                      |¦  «        ¦  «        ¦  «        z   }|S ©zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        ©rs   r6   rq   rv   ru   rt   ©r=   rC   s     rA   rY   zOutlooker.forward³   óa   € ð �—’ §	¢	¨$¯*ª*°Q©-¬-Ñ 8Ô 8Ñ9Ô9Ñ9ˆØ�—’ §¢¨¯ª°A©¬Ñ 7Ô 7Ñ8Ô8Ñ8ˆØˆrB   )rZ   r[   r\   r]   r3   ÚGELUÚ	LayerNormr^   r`   r
   ÚModuler_   r1   ra   rb   rY   rc   rd   s   @rA   rf   rf   y   s  ø€ € € € € ØCÐCð ØØ!Ø!Ø!Ø)+¬Ø*,¬,Ø"ØØð5Sð 5Sàð5Sð ð5Sð ð	5Sð
 ð5Sð ð5Sð ð5Sð ð5Sð ð5Sð ˜BœI”ð5Sð ˜RœYœð5Sð ð5Sð 5Sð 5Sð 5Sð 5Sð 5Sðn˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   rf   c                   ó–   ‡ — e Zd ZU dZej        j        e         ed<   	 	 	 	 	 	 dde	de	d	ed
e
de
f
ˆ fd„Zdej        dej        fd„Zˆ xZS )Ú	Attentionz!Multi-head self-attention module.Ú
fused_attné   Fr   Nr    r!   r%   r&   r'   c                 ór  •— ||dœ}t          ¦   «                              ¦   «          || _        ||z  }	|	dz  | _        t	          ¦   «         | _        t          j        ||dz  fd|i|¤Ž| _        t          j	        |¦  «        | _
        t          j        ||fi |¤Ž| _        t          j	        |¦  «        | _        dS )a,  Initialize Attention module.

        Args:
            dim: Input feature dimension.
            num_heads: Number of attention heads.
            qkv_bias: Whether to use bias in QKV projection.
            attn_drop: Attention dropout rate.
            proj_drop: Projection dropout rate.
        r)   r,   r   r-   N)r0   r1   r!   r2   r   r�   r3   r4   Úqkvr7   r&   r8   r'   )r=   r    r!   r%   r&   r'   r*   r+   r>   r?   r@   s             €rA   r1   zAttention.__init__Å   s·   ø€ ð& ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ"ˆŒØ˜)Ñ#ˆØ Ñ%ˆŒ
Ý(Ñ*Ô*ˆŒå”9˜S #¨¡'Ð?Ð?°Ð?¸BÐ?Ð?ˆŒÝœ IÑ.Ô.ˆŒÝ”I˜c 3Ð-Ð-¨"Ð-Ð-ˆŒ	Ýœ IÑ.Ô.ˆŒˆˆrB   rC   rD   c                 ó®  — |j         \  }}}}|                      |¦  «                             |||z  d| j        || j        z  ¦  «                             ddddd¦  «        }|                     d¦  «        \  }}}	| j        r,t          j        |||	| j	        r| j
        j        nd¬¦  «        }nS|| j        z  }||                     dd	¦  «        z  }
|
                     d	¬
¦  «        }
|  
                    |
¦  «        }
|
|	z  }|                     dd¦  «                             ||||¦  «        }|                      |¦  «        }|                      |¦  «        }|S )rF   r   rG   r   r   r.   r   )Ú	dropout_péþÿÿÿrH   rI   )rK   r„   rO   r!   rL   Úunbindr�   rQ   Úscaled_dot_product_attentionÚtrainingr&   Úpr2   Ú	transposerP   r8   r'   )r=   rC   rS   rT   rU   rV   r„   ÚqÚkr5   r6   s              rA   rY   zAttention.forwardä   sQ  € ð ”W‰
ˆˆ1ˆa�à�hŠh�q‰kŒk×!Ò! ! Q¨¡U¨A¨t¬~¸qÀDÄNÑ?RÑSÔS×[Ò[Ð\]Ð_`ÐbcÐefÐhiÑjÔjˆØ—*’*˜Q‘-”-‰ˆˆ1ˆaàŒ?ð 
	ÝÔ.Ø�1�aØ.2¬mÐC˜$œ.Ô*Ð*Àðñ ô ˆAˆAð
 �D”J‘ˆAØ�q—{’{ 2 rÑ*Ô*Ñ*ˆDØ—<’< B�<Ñ'Ô'ˆDØ—>’> $Ñ'Ô'ˆDØ�q‘ˆAà�KŠK˜˜1ÑÔ×%Ò% a¨¨A¨qÑ1Ô1ˆØ�IŠI�a‰LŒLˆØ�NŠN˜1ÑÔˆàˆrB   )r‚   Fr   r   NN)rZ   r[   r\   r]   ra   ÚjitÚFinalr_   Ú__annotations__r^   r`   r1   rb   rY   rc   rd   s   @rA   r€   r€   Á   sÉ   ø€ € € € € € Ø+Ð+Ø”	” Ô%Ð%Ð%Ñ%ð
 Ø"Ø!Ø!ØØð/ð /àð/ð ð/ð ð	/ð
 ð/ð ð/ð /ð /ð /ð /ð /ð>˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   r€   c                   óº   ‡ — e Zd ZdZddddej        ej        ddfdededed	e	d
edede
ej                 de
ej                 fˆ fd„Zdej        dej        fd„Zˆ xZS )ÚTransformerz9Transformer block with multi-head self-attention and MLP.ç      @Fr   Nr    r!   rh   r%   r&   ri   rj   rk   c                 ó´  •— |	|
dœ}t          ¦   «                              ¦   «           ||fi |¤Ž| _        t          |f|||dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _         ||fi |¤Ž| _	        t          d|t          ||z  ¦  «        |dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _        dS )aÇ  Initialize Transformer block.

        Args:
            dim: Input feature dimension.
            num_heads: Number of attention heads.
            mlp_ratio: Ratio for MLP hidden dimension.
            qkv_bias: Whether to use bias in QKV projection.
            attn_drop: Attention dropout rate.
            drop_path: Stochastic depth drop rate.
            act_layer: Activation layer type.
            norm_layer: Normalization layer type.
        r)   )r!   r%   r&   r   rm   Nrp   )r0   r1   rq   r€   r6   r   r3   rr   rs   rt   r   r^   ru   rv   )r=   r    r!   rh   r%   r&   ri   rj   rk   r*   r+   r>   r@   s               €rA   r1   zTransformer.__init__  sð   ø€ ð2 ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ�Z Ð*Ð* rÐ*Ð*ˆŒ
Ý˜cÐe¨YÀÐU^ÐeÐeÐbdÐeÐeˆŒ	Ø1:¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒà�Z Ð*Ð* rÐ*Ð*ˆŒ
ÝÐh 3½¸CÀ)¹OÑ8LÔ8LÐXaÐhÐhÐegÐhÐhˆŒØ1:¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒˆˆrB   rC   rD   c                 óþ   — ||                       |                      |                      |¦  «        ¦  «        ¦  «        z   }||                      |                      |                      |¦  «        ¦  «        ¦  «        z   }|S rx   ry   rz   s     rA   rY   zTransformer.forward+  r{   rB   )rZ   r[   r\   r]   r3   r|   r}   r^   r`   r_   r
   r~   r1   ra   rb   rY   rc   rd   s   @rA   r“   r“     sö   ø€ € € € € ØCÐCð  "Ø"Ø!Ø!Ø)+¬Ø*,¬,ØØð!Sð !Sàð!Sð ð!Sð ð	!Sð
 ð!Sð ð!Sð ð!Sð ˜BœI”ð!Sð ˜RœYœð!Sð !Sð !Sð !Sð !Sð !SðF˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   r“   c                   ó|   ‡ — e Zd ZdZ	 	 	 	 	 	 	 ddededee         d	ed
edefˆ fd„Zde	j
        de	j
        fd„Zˆ xZS )ÚClassAttentionz6Class attention mechanism for class token interaction.r‚   NFr   r    r!   r?   r%   r&   r'   c	                 óð  •— ||dœ}	t          ¦   «                              ¦   «          || _        |�|| _        n||z  }|| _        |dz  | _        t          j        || j        | j        z  dz  fd|i|	¤Ž| _        t          j        || j        | j        z  fd|i|	¤Ž| _        t          j	        |¦  «        | _
        t          j        | j        | j        z  |fi |	¤Ž| _        t          j	        |¦  «        | _        dS )a{  Initialize ClassAttention.

        Args:
            dim: Input feature dimension.
            num_heads: Number of attention heads.
            head_dim: Dimension per head. If None, computed as dim // num_heads.
            qkv_bias: Whether to use bias in QKV projection.
            attn_drop: Attention dropout rate.
            proj_drop: Projection dropout rate.
        r)   Nr,   rG   r-   )r0   r1   r!   r?   r2   r3   r4   Úkvr�   r7   r&   r8   r'   )r=   r    r!   r?   r%   r&   r'   r*   r+   r>   r@   s             €rA   r1   zClassAttention.__init__<  sþ   ø€ ð* ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ"ˆŒØÐØ$ˆDŒMˆMà˜iÑ'ˆHØ$ˆDŒMØ Ñ%ˆŒ
å”)˜C ¤°´Ñ!?À!Ñ!CÐYÐYÈ(ÐYÐVXÐYÐYˆŒÝ”˜3 ¤°´Ñ >ÐTÐTÀXÐTÐQSÐTÐTˆŒÝœ IÑ.Ô.ˆŒÝ”I˜dœm¨d¬nÑ<¸cÐHÐHÀRÐHÐHˆŒ	Ýœ IÑ.Ô.ˆŒˆˆrB   rC   rD   c                 óÐ  — |j         \  }}}|                      |¦  «                             ||d| j        | j        ¦  «                             ddddd¦  «        }|                     d¦  «        \  }}|                      |dd…dd…dd…f         ¦  «                             || j        d| j        ¦  «        | j        z  }|| 	                    dd¦  «        z  }	|	 
                    d¬	¦  «        }	|                      |	¦  «        }	|	|z   	                    dd¦  «                             |d| j        | j        z  ¦  «        }
|                      |
¦  «        }
|                      |
¦  «        }
|
S )
z»Forward pass.

        Args:
            x: Input tensor of shape (B, N, C) where first token is class token.

        Returns:
            Class token output of shape (B, 1, C).
        rG   r   r   r   r.   Nr‡   rH   rI   )rK   rš   rO   r!   r?   rL   rˆ   r�   r2   rŒ   rP   r&   r8   r'   )r=   rC   rS   ÚNrV   rš   rŽ   r5   r�   r6   Ú	cls_embeds              rA   rY   zClassAttention.forwarda  sF  € ð ”'‰ˆˆ1ˆaà�WŠW�Q‰ZŒZ×Ò  1 a¨¬¸¼ÑGÔG×OÒOÐPQÐSTÐVWÐYZÐ\]Ñ^Ô^ˆØ�yŠy˜‰|Œ|‰ˆˆ1Ø�FŠF�1�Q�Q�Q˜˜˜˜A˜A˜A�X”;ÑÔ×'Ò'¨¨4¬>¸1¸d¼mÑLÔLÈtÌzÑYˆà�1—;’;˜r 2Ñ&Ô&Ñ&ˆØ�|Š| ˆ|Ñ#Ô#ˆØ�~Š~˜dÑ#Ô#ˆà˜A‘X×(Ò(¨¨AÑ.Ô.×6Ò6°q¸!¸T¼]ÈTÌ^Ñ=[Ñ\Ô\ˆ	Ø—I’I˜iÑ(Ô(ˆ	Ø—N’N 9Ñ-Ô-ˆ	ØÐrB   )r‚   NFr   r   NN)rZ   r[   r\   r]   r^   r   r_   r`   r1   ra   rb   rY   rc   rd   s   @rA   r˜   r˜   9  sÄ   ø€ € € € € Ø@Ð@ð
 Ø&*Ø"Ø!Ø!ØØð#/ð #/àð#/ð ð#/ð ˜s”mð	#/ð
 ð#/ð ð#/ð ð#/ð #/ð #/ð #/ð #/ð #/ðJ˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   r˜   c                   óÒ   ‡ — e Zd ZdZddddddej        ej        ddf
dededee         d	e	d
e
de	de	de	deej                 deej                 fˆ fd„Zdej        dej        fd„Zˆ xZS )Ú
ClassBlockz3Class block that combines class attention with MLP.Nr”   Fr   r    r!   r?   rh   r%   Údropr&   ri   rj   rk   c           	      óº  •— ||dœ}t          ¦   «                              ¦   «           |
|fi |¤Ž| _        t          |f|||||dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _         |
|fi |¤Ž| _	        t          d|t          ||z  ¦  «        |	|dœ|¤Ž| _        |dk    rt          |¦  «        nt          j        ¦   «         | _        dS )a1  Initialize ClassBlock.

        Args:
            dim: Input feature dimension.
            num_heads: Number of attention heads.
            head_dim: Dimension per head. If None, computed as dim // num_heads.
            mlp_ratio: Ratio for MLP hidden dimension.
            qkv_bias: Whether to use bias in QKV projection.
            drop: Dropout rate.
            attn_drop: Attention dropout rate.
            drop_path: Stochastic depth drop rate.
            act_layer: Activation layer type.
            norm_layer: Normalization layer type.
        r)   )r!   r?   r%   r&   r'   r   )rn   ro   rj   r    Nrp   )r0   r1   rq   r˜   r6   r   r3   rr   rs   rt   r   r^   ru   rv   )r=   r    r!   r?   rh   r%   r    r&   ri   rj   rk   r*   r+   r>   r@   s                 €rA   r1   zClassBlock.__init__}  s  ø€ ð: ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ�Z Ð*Ð* rÐ*Ð*ˆŒ
Ý"Øð
àØØØØð
ð 
ð ð
ð 
ˆŒ	ð 2;¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒà�Z Ð*Ð* rÐ*Ð*ˆŒ
Ýð 
ØÝ  i¡Ñ0Ô0ØØð	
ð 
ð
 ð
ð 
ˆŒð 2;¸R²°�( 9Ñ-Ô-Ð-ÅRÄ[Á]Ä]ˆŒˆˆrB   rC   rD   c                 ó^  — |dd…dd…f         }||                       |                      |                      |¦  «        ¦  «        ¦  «        z   }||                      |                      |                      |¦  «        ¦  «        ¦  «        z   }t          j        ||dd…dd…f         gd¬¦  «        S )z¼Forward pass.

        Args:
            x: Input tensor of shape (B, N, C) where first token is class token.

        Returns:
            Output tensor with updated class token.
        Nr   rI   )rs   r6   rq   rv   ru   rt   ra   Úcat)r=   rC   r�   s      rA   rY   zClassBlock.forward²  s�   € ð �a�a�a˜˜!˜�e”Hˆ	Ø §¢°·	²	¸$¿*º*ÀQ¹-¼-Ñ0HÔ0HÑ IÔ IÑIˆ	Ø §¢°·²¸¿ºÀIÑ9NÔ9NÑ0OÔ0OÑ PÔ PÑPˆ	ÝŒy˜) Q q q q¨!¨"¨" u¤XÐ.°AÐ6Ñ6Ô6Ð6rB   )rZ   r[   r\   r]   r3   r|   r}   r^   r   r`   r_   r
   r~   r1   ra   rb   rY   rc   rd   s   @rA   rŸ   rŸ   z  s  ø€ € € € € Ø=Ð=ð '+Ø!Ø"ØØ!Ø!Ø)+¬Ø*,¬,ØØð3Sð 3Sàð3Sð ð3Sð ˜s”mð	3Sð
 ð3Sð ð3Sð ð3Sð ð3Sð ð3Sð ˜BœI”ð3Sð ˜RœYœð3Sð 3Sð 3Sð 3Sð 3Sð 3Sðj7˜œð 7¨%¬,ð 7ð 7ð 7ð 7ð 7ð 7ð 7ð 7rB   rŸ   Ú
block_typeÚkwargsrD   c                 ó@   — | dk    rt          di |¤ŽS J d| › �¦   «         ‚)zÍGet block based on type.

    Args:
        block_type: Type of block ('ca' for ClassBlock).
        **kwargs: Additional keyword arguments for block.

    Returns:
        The requested block module.
    ÚcaFzInvalid block type: Nrp   )rŸ   )r¤   r¥   s     rA   Ú	get_blockr¨   Á  s9   € ð �TÒÐÝÐ#Ð#˜FÐ#Ð#Ð#à9Ð9¨ZÐ9Ð9Ñ9Ô9ˆurB   Úsize.Úlamr2   c                 ó  — | d         |z  }| d         |z  }t          j        |t           j        ¬¦  «        }t          j        |t           j        ¬¦  «        }t          j        d|z
  ¦  «        }||z                       ¦   «         }||z                       ¦   «         }	t          j        d|d¦  «        }
t          j        d|d¦  «        }t          j        |
|dz  z
  d|¦  «        }t          j        ||	dz  z
  d|¦  «        }t          j        |
|dz   dz  z   d|¦  «        }t          j        ||	dz   dz  z   d|¦  «        }|                     ¦   «         |                     ¦   «         |                     ¦   «         |                     ¦   «         fS )a1  Get random bounding box for token labeling.

    Reference: https://github.com/zihangJiang/TokenLabeling

    Args:
        size: Input tensor size tuple.
        lam: Lambda parameter for cutmix.
        scale: Scaling factor.

    Returns:
        Bounding box coordinates (bbx1, bby1, bbx2, bby2).
    r   rG   )r+   ç      ð?r   ©r   )ra   ÚtensorÚfloat32Úsqrtr^   ÚrandintÚclampÚitem)r©   rª   r2   rU   rT   ÚW_tÚH_tÚcut_ratÚcut_wÚcut_hÚcxÚcyÚbbx1Úbby1Úbbx2Úbby2s                   rA   Ú	rand_bboxr¿   Ñ  sX  € ð 	ˆQŒ�5Ñ€AØˆQŒ�5Ñ€AÝ
Œ,�q¥¤Ð
.Ñ
.Ô
.€CÝ
Œ,�q¥¤Ð
.Ñ
.Ô
.€CÝŒj˜˜c™Ñ"Ô"€GØ�7‰]×ÒÑ!Ô!€EØ�7‰]×ÒÑ!Ô!€Eõ 
Œ�q˜!˜TÑ	"Ô	"€BÝ	Œ�q˜!˜TÑ	"Ô	"€BåŒ;�r˜E Q™J‘¨¨1Ñ-Ô-€DÝŒ;�r˜E Q™J‘¨¨1Ñ-Ô-€DÝŒ;�r˜U Q™Y¨1Ñ,Ñ,¨a°Ñ3Ô3€DÝŒ;�r˜U Q™Y¨1Ñ,Ñ,¨a°Ñ3Ô3€Dà�9Š9‰;Œ;˜Ÿ	š	™œ T§Y¢Y¡[¤[°$·)²)±+´+Ð=Ð=rB   c                   óx   ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 	 dd
ededededededefˆ fd„Zdej        dej        fd„Z	ˆ xZ
S )Ú
PatchEmbedz6Image to patch embedding with multi-layer convolution.éà   Fr   r‚   r   é@   é€  NÚimg_sizeÚ	stem_convÚstem_strideÚ
patch_sizeÚin_chansÚ
hidden_dimÚ	embed_dimc
                 ól  •— ||	dœ}
t          ¦   «                              ¦   «          |dv sJ ‚|rÍt          j        t          j        ||fd|dddœ|
¤Žt          j        |fi |
¤Žt          j        d¬¦  «        t          j        ||fdd	d	ddœ|
¤Žt          j        |fi |
¤Žt          j        d¬¦  «        t          j        ||fdd	d	ddœ|
¤Žt          j        |fi |
¤Žt          j        d¬¦  «        ¦	  «	        | _        nd
| _        t          j        ||f||z  ||z  dœ|
¤Ž| _        ||z  ||z  z  | _	        d
S )a  Initialize PatchEmbed.

        Different from ViT which uses 1 conv layer, VOLO uses multiple conv layers for patch embedding.

        Args:
            img_size: Input image size.
            stem_conv: Whether to use stem convolution layers.
            stem_stride: Stride for stem convolution.
            patch_size: Patch size (must be 4, 8, or 16).
            in_chans: Number of input channels.
            hidden_dim: Hidden dimension for stem convolution.
            embed_dim: Output embedding dimension.
        r)   )r.   r‚   é   é   r   F)r"   r$   r#   r-   T)Úinplacer   N©r"   r$   )
r0   r1   r3   Ú
SequentialÚConv2dÚBatchNorm2dÚReLUÚconvr8   Únum_patches)r=   rÅ   rÆ   rÇ   rÈ   rÉ   rÊ   rË   r*   r+   r>   r@   s              €rA   r1   zPatchEmbed.__init__õ  s›  ø€ ð2 ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ˜ZÐ'Ð'Ð'Ð'Øð 	ÝœÝ”	˜( JÐo¸AÀkÐ[\ÐchÐoÐoÐlnÐoÐoÝ”˜zÐ0Ð0¨RÐ0Ð0Ý” Ð%Ñ%Ô%Ý”	˜* jÐg¸aÈÐSTÐ[`ÐgÐgÐdfÐgÐgÝ”˜zÐ0Ð0¨RÐ0Ð0Ý” Ð%Ñ%Ô%Ý”	˜* jÐg¸aÈÐSTÐ[`ÐgÐgÐdfÐgÐgÝ”˜zÐ0Ð0¨RÐ0Ð0Ý” Ð%Ñ%Ô%ñ
ô 
ˆDŒIˆIð ˆDŒIå”IØØð
ð # kÑ1Ø Ñ,ð	
ð 
ð
 ð
ð 
ˆŒ	ð %¨
Ñ2°xÀ:Ñ7MÑNˆÔÐÐrB   rC   rD   c                 óh   — | j         �|                       |¦  «        }|                      |¦  «        }|S )z¥Forward pass.

        Args:
            x: Input tensor of shape (B, C, H, W).

        Returns:
            Output tensor of shape (B, embed_dim, H', W').
        )rÕ   r8   rz   s     rA   rY   zPatchEmbed.forward)  s/   € ð Œ9Ð Ø—	’	˜!‘”ˆAØ�IŠI�a‰LŒLˆØˆrB   )	rÂ   Fr   r‚   r   rÃ   rÄ   NN)rZ   r[   r\   r]   r^   r_   r1   ra   rb   rY   rc   rd   s   @rA   rÁ   rÁ   ò  sÞ   ø€ € € € € Ø@Ð@ð  Ø#Ø ØØØ Ø ØØð2Oð 2Oàð2Oð ð2Oð ð	2Oð
 ð2Oð ð2Oð ð2Oð ð2Oð 2Oð 2Oð 2Oð 2Oð 2Oðh˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   rÁ   c                   ó\   ‡ — e Zd ZdZ	 	 	 ddededefˆ fd„Zdej        d	ej        fd
„Zˆ xZ	S )Ú
Downsamplez#Downsampling module between stages.rG   NÚin_embed_dimÚout_embed_dimrÈ   c                 óˆ   •— t          ¦   «                              ¦   «          ||dœ}t          j        ||f||dœ|¤Ž| _        dS )zÏInitialize Downsample.

        Args:
            in_embed_dim: Input embedding dimension.
            out_embed_dim: Output embedding dimension.
            patch_size: Patch size for downsampling.
        r)   rÐ   N)r0   r1   r3   rÒ   r8   )r=   rÚ   rÛ   rÈ   r*   r+   r>   r@   s          €rA   r1   zDownsample.__init__;  sP   ø€ õ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆÝ”I˜l¨MÐkÀzÐZdÐkÐkÐhjÐkÐkˆŒ	ˆ	ˆ	rB   rC   rD   c                 ó�   — |                      dddd¦  «        }|                      |¦  «        }|                      dddd¦  «        }|S )zžForward pass.

        Args:
            x: Input tensor of shape (B, H, W, C).

        Returns:
            Output tensor of shape (B, H', W', C').
        r   r   r   rG   )rL   r8   rz   s     rA   rY   zDownsample.forwardN  sF   € ð �IŠI�a˜˜A˜qÑ!Ô!ˆØ�IŠI�a‰LŒLˆØ�IŠI�a˜˜A˜qÑ!Ô!ˆØˆrB   )rG   NN)
rZ   r[   r\   r]   r^   r1   ra   rb   rY   rc   rd   s   @rA   rÙ   rÙ   8  sŸ   ø€ € € € € Ø-Ð-ð  ØØðlð làðlð ðlð ð	lð lð lð lð lð lð&˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   rÙ   r   rG   rg   Fr   Úblock_fnÚindexr    Úlayersr!   r"   r#   r$   rh   r%   r&   Údrop_path_ratec                 ó  — g }t          ||         ¦  «        D ]Y}||t          |d|…         ¦  «        z   z  t          |¦  «        dz
  z  }|                      | |f||||||	|
|||dœ
|¤Ž¦  «         ŒZt          j        |Ž }|S )aõ  Generate outlooker layers for stage 1.

    Args:
        block_fn: Block function to use (typically Outlooker).
        index: Index of current stage.
        dim: Feature dimension.
        layers: List of layer counts for each stage.
        num_heads: Number of attention heads.
        kernel_size: Kernel size for outlook attention.
        padding: Padding for outlook attention.
        stride: Stride for outlook attention.
        mlp_ratio: Ratio for MLP hidden dimension.
        qkv_bias: Whether to use bias in QKV projection.
        attn_drop: Attention dropout rate.
        drop_path_rate: Stochastic depth drop rate.
        **kwargs: Additional keyword arguments.

    Returns:
        Sequential module containing outlooker blocks.
    Nr   )
r"   r#   r$   r!   rh   r%   r&   ri   r*   r+   ©ÚrangeÚsumÚappendr3   rÑ   )rÞ   rß   r    rà   r!   r"   r#   r$   rh   r%   r&   rá   r*   r+   r¥   ÚblocksÚ	block_idxÚ	block_dprs                     rA   Úoutlooker_blocksrê   ]  sÅ   € ðJ €FÝ˜6 %œ=Ñ)Ô)ð ð ˆ	Ø" iµ#°f¸V¸e¸V´nÑ2EÔ2EÑ&EÑFÍ#ÈfÉ+Ì+ÐXYÉ/ÑZˆ	Ø�Š�h�hØð
à#ØØØØØØØØØð
ð 
ð ð
ð 
ñ 	ô 	ð 	ð 	õ Œ]˜FÐ#€FØ€MrB   c	                 óú   — g }
t          ||         ¦  «        D ]T}||t          |d|…         ¦  «        z   z  t          |¦  «        dz
  z  }|
                      | ||f||||dœ|	¤Ž¦  «         ŒUt          j        |
Ž }
|
S )ae  Generate transformer layers for stage 2.

    Args:
        block_fn: Block function to use (typically Transformer).
        index: Index of current stage.
        dim: Feature dimension.
        layers: List of layer counts for each stage.
        num_heads: Number of attention heads.
        mlp_ratio: Ratio for MLP hidden dimension.
        qkv_bias: Whether to use bias in QKV projection.
        attn_drop: Attention dropout rate.
        drop_path_rate: Stochastic depth drop rate.
        **kwargs: Additional keyword arguments.

    Returns:
        Sequential module containing transformer blocks.
    Nr   )rh   r%   r&   ri   rã   )rÞ   rß   r    rà   r!   rh   r%   r&   rá   r¥   rç   rè   ré   s                rA   Útransformer_blocksrì   —  s·   € ð: €FÝ˜6 %œ=Ñ)Ô)ð 
ð 
ˆ	Ø" iµ#°f¸V¸e¸V´nÑ2EÔ2EÑ&EÑFÍ#ÈfÉ+Ì+ÐXYÉ/ÑZˆ	Ø�Š�h�hØØð
ð  ØØØð
ð 
ð ð
ð 
ñ 	ô 	ð 	ð 	õ Œ]˜FÐ#€FØ€MrB   c            -       óŽ  ‡ — e Zd ZdZddddddddd	d	d
dddddej        ddddddfdee         dedededededede	ee                  de	ee                  de
edf         de
edf         dedededed ed!ed"eej                 d#e	e
edf                  d$ed%ed&ef,ˆ fd'„Zd(ej        d)dfd*„Zej        j        d)efd+„¦   «         Zej        j        dEd,ed)eeef         fd-„¦   «         Zej        j        dFd.ed)dfd/„¦   «         Zej        j        d)ej        fd0„¦   «         ZdGdede	e         d)dfd1„Zd2ej        d)ej        fd3„Zd2ej        d)ej        fd4„Zd2ej        d)eej        e
ej        ej        e
eeeef         f         f         fd5„Z 	 	 	 	 	 dHd2ej        d7e	eeee         f                  d8ed9ed:ed;ed)eeej                 e
ej        eej                 f         f         fd<„Z!	 	 	 dId7eeee         f         d>ed?ed)ee         fd@„Z"d2ej        d)ej        fdA„Z#dEd2ej        dBed)ej        fdC„Z$d2ej        d)ej        fdD„Z%ˆ xZ&S )Jr   zVision Outlooker (VOLO) model.rÂ   r   éè  Útokenr‚   rÃ   N)TFFFrg   Fr   )r§   r§   TrG   rà   rÅ   rÉ   Únum_classesÚglobal_poolrÈ   Ústem_hidden_dimÚ
embed_dimsr!   Údownsamples.Úoutlook_attentionrh   r%   Ú	drop_rateÚpos_drop_rateÚattn_drop_raterá   rk   Úpost_layersÚuse_aux_headÚuse_mix_tokenÚpooling_scalec           
      ó(  •‡‡	‡‡‡‡‡‡ — t          ¦   «                              ¦   «          ||dœŠ t          |¦  «        } t          |¦  «        ‰¦  «        Št	          |¦  «        }|| _        || _        || _        || _        || _	        ‰d         x| _
        | _        |rd| _        |dk    s
J d¦   «         ‚d| _        t          ddd|||‰d	         d
œ‰ ¤Ž| _        |}|d	         |z  |z  |d         |z  |z  f}t!          j        t%          j        d|d	         |d         ‰d         fi ‰ ¤Ž¦  «        | _        t!          j        |¬¦  «        | _        g | _        g | _        g }d	}t3          t          |¦  «        ¦  «        D �]}||         r.t5          t6          |‰|         |‰	|         f‰|         ‰‰‰dœ‰ ¤Ž}n.t9          t:          |‰|         |‰	|         f‰|         ‰|‰‰dœ‰ ¤Ž}|                     |¦  «         | j                             |¦  «         | j                             t?          ‰|         |d|› �¬¦  «        ¦  «         |dz  }|
|         r;|                     tA          ‰|         ‰|dz            dfi ‰ ¤Ž¦  «         |dz  }|dz  }�Œt!          j!        |¦  «        | _"        d| _#        ‰��t!          j!        ˆˆ ˆˆˆˆ	ˆˆfd„t3          t          ‰¦  «        ¦  «        D ¦   «         ¦  «        | _#        t!          j        t%          j        dd‰d         fi ‰ ¤Ž¦  «        | _$        tK          | j$        d¬¦  «         |r7|d	k    rt!          j&        | j
        |fi ‰ ¤Žnt!          j'        ¦   «         | _(        nd| _(         ‰| j
        fi ‰ ¤Ž| _)        t!          j        |¦  «        | _*        |d	k    rt!          j&        | j
        |fi ‰ ¤Žnt!          j'        ¦   «         | _+        tK          | j        d¬¦  «         |  ,                    | j-        ¦  «         dS )a/  Initialize VOLO model.

        Args:
            layers: Number of blocks in each stage.
            img_size: Input image size.
            in_chans: Number of input channels.
            num_classes: Number of classes for classification.
            global_pool: Global pooling type ('token', 'avg', or '').
            patch_size: Patch size for patch embedding.
            stem_hidden_dim: Hidden dimension for stem convolution.
            embed_dims: List of embedding dimensions for each stage.
            num_heads: List of number of attention heads for each stage.
            downsamples: Whether to downsample between stages.
            outlook_attention: Whether to use outlook attention in each stage.
            mlp_ratio: Ratio for MLP hidden dimension.
            qkv_bias: Whether to use bias in QKV projection.
            drop_rate: Dropout rate.
            pos_drop_rate: Position embedding dropout rate.
            attn_drop_rate: Attention dropout rate.
            drop_path_rate: Stochastic depth drop rate.
            norm_layer: Normalization layer type.
            post_layers: Post-processing layer types.
            use_aux_head: Whether to use auxiliary head.
            use_mix_token: Whether to use token mixing for training.
            pooling_scale: Pooling scale factor.
        r)   rH   r¬   rï   z)return all tokens if mix_token is enabledFTrG   r   )rÆ   rÇ   rÈ   rÉ   rÊ   rË   r   )r‹   )rh   r%   r&   rk   )rh   r%   rá   r&   rk   znetwork.)Únum_chsÚ	reductionÚmoduleNc                 ój   •— g | ]/}t          ‰|         f‰d          ‰d          ‰d          ‰	‰d‰dœ‰¤Ž‘Œ0S )rH   r   )r    r!   rh   r%   r&   ri   rk   )r¨   )
Ú.0Úirø   r>   ró   rh   rk   r!   rù   r%   s
     €€€€€€€€rA   ú
<listcomp>z!VOLO.__init__.<locals>.<listcomp>O  sq   ø€ ð /ð /ð /ð õ Ø ”Nð
à" 2œØ'¨œmØ'¨œmØ%Ø,Ø Ø)ð
ð 
ð ð
ð 
ð/ð /ð /rB   ç{®Gáz”?©Ústdrp   ).r0   r1   Úlenr   r   rð   rÉ   rñ   Ú	mix_tokenrü   Únum_featuresÚhead_hidden_sizeÚbetaÚgrad_checkpointingrÁ   Úpatch_embedr3   Ú	Parameterra   ÚzerosÚ	pos_embedr7   Úpos_dropÚ
stage_endsÚfeature_inforä   rê   rf   rì   r“   ræ   ÚdictrÙ   Ú
ModuleListÚnetworkÚpost_networkÚ	cls_tokenr   r4   rr   Úaux_headÚnormÚ	head_dropÚheadÚapplyÚ_init_weights)"r=   rà   rÅ   rÉ   rð   rñ   rÈ   rò   ró   r!   rô   rõ   rh   r%   rö   r÷   rø   rá   rk   rù   rú   rû   rü   r*   r+   Ú
num_layersÚrÚ
patch_gridr  rè   r  Ústager>   r@   s"           ``  ``  ` ``            @€rA   r1   zVOLO.__init__Ç  sÔ  øøøøøøøøø€ õj 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆÝ˜‘[”[ˆ
Ø)•I˜jÑ)Ô)¨)Ñ4Ô4ˆ	Ý˜XÑ&Ô&ˆà&ˆÔØ ˆŒØ&ˆÔØ&ˆŒØ*ˆÔØ4>¸r´NÐBˆÔ˜DÔ1Øð 	WØˆDŒIØ 'Ò)Ð)Ð)Ð+VÑ)Ô)Ð)Ø"'ˆÔå%ð 
ØØØ!ØØ&Ø  ”mð
ð 
ð ð
ð 
ˆÔð ˆð ˜q”k ZÑ/°=Ñ@À(È1Ä+ÐQ[ÑB[Ð_lÑBlÐmˆ
Ýœ¥e¤k°!°ZÀ´]ÀJÈqÄMÐS]Ð^`ÔSaÐ&hÐ&hÐegÐ&hÐ&hÑiÔiˆŒÝœ
 ]Ð3Ñ3Ô3ˆŒð ˆŒØˆÔØˆØˆ	Ý•s˜6‘{”{Ñ#Ô#ð &	ñ &	ˆAØ  Ô#ð å(ÝØØ˜q”MØØ˜a”Lðð (¨œlØ%Ø,Ø)ðð ð ðð ��õ +ÝØØ˜q”MØØ˜a”Lðð (¨œlØ%Ø#1Ø,Ø)ðð ð ðð �ð �NŠN˜5Ñ!Ô!Ð!ØŒO×"Ò" 9Ñ-Ô-Ð-ØÔ×$Ò$¥T°*¸Q´-È1ÐUkÐ`iÐUkÐUkÐ%lÑ%lÔ%lÑmÔmÐmØ˜‰NˆIØ˜1Œ~ð à—’�z¨*°Q¬-¸ÀAÈÁEÔ9JÈAÐTÐTÐQSÐTÐTÑUÔUÐUØ�Q‘�Ø˜Q‘�	ùå”} WÑ-Ô-ˆŒð !ˆÔØÐ"Ý "¤ð /ð /ð /ð /ð /ð /ð /ð /ð /ð /ð /õ �s ;Ñ/Ô/Ñ0Ô0ð/ñ /ô /ñ !ô !ˆDÔõ  œ\­%¬+°a¸¸JÀr¼NÐ*QÐ*QÈbÐ*QÐ*QÑRÔRˆDŒNÝ˜$œ.¨cÐ2Ñ2Ô2Ð2ð ð 	!ØOZÐ]^ÊÈ�BœI dÔ&7¸ÐKÐKÈÐKÐKÐKÕdfÔdoÑdqÔdqˆDŒMˆMà ˆDŒMØ�J˜tÔ0Ð7Ð7°BÐ7Ð7ˆŒ	õ œ IÑ.Ô.ˆŒØGRÐUVÂÀ•B”I˜dÔ/°ÐCÐCÀÐCÐCÐCÕ\^Ô\gÑ\iÔ\iˆŒ	å�d”n¨#Ð.Ñ.Ô.Ð.Ø�
Š
�4Ô%Ñ&Ô&Ð&Ð&Ð&rB   ÚmrD   c                 óþ   — t          |t          j        ¦  «        r^t          |j        d¬¦  «         t          |t          j        ¦  «        r0|j        �+t          j                             |j        d¦  «         dS dS dS dS )z\Initialize weights for modules.

        Args:
            m: Module to initialize.
        r  r  Nr   )Ú
isinstancer3   r4   r   Úweightr-   ÚinitÚ	constant_)r=   r$  s     rA   r  zVOLO._init_weightsn  s‚   € õ �a�œÑ#Ô#ð 	-Ý˜!œ(¨Ð,Ñ,Ô,Ð,Ý˜!�RœYÑ'Ô'ð -¨A¬FÐ,>Ý”×!Ò! !¤&¨!Ñ,Ô,Ð,Ð,Ð,ð	-ð 	-ð-ð -Ð,>Ð,>rB   c                 ó
   — ddhS )zwGet set of parameters that should not have weight decay.

        Returns:
            Set of parameter names.
        r  r  rp   ©r=   s    rA   Úno_weight_decayzVOLO.no_weight_decayy  s   € ð ˜[Ð)Ð)rB   Úcoarsec                 ó.   — t          dddgg d¢¬¦  «        S )z®Get parameter grouping for optimizer.

        Args:
            coarse: Whether to use coarse grouping.

        Returns:
            Parameter grouping dictionary.
        z ^cls_token|pos_embed|patch_embed)z^network\.(\d+)\.(\d+)N)z^network\.(\d+)©r   ))z
^cls_tokenr/  )z^post_network\.(\d+)N)z^norm)iŸ† )Ústemrç   Úblocks2)r  )r=   r-  s     rA   Úgroup_matcherzVOLO.group_matcher‚  s8   € õ Ø4à1Ø*ððð ð ð
ñ 
ô 
ð 	
rB   Úenablec                 ó   — || _         dS )zqSet gradient checkpointing.

        Args:
            enable: Whether to enable gradient checkpointing.
        N)r  )r=   r3  s     rA   Úset_grad_checkpointingzVOLO.set_grad_checkpointing™  s   € ð #)ˆÔÐÐrB   c                 ó   — | j         S )zYGet classifier module.

        Returns:
            The classifier head module.
        )r  r+  s    rA   Úget_classifierzVOLO.get_classifier¢  s   € ð ŒyÐrB   c                 óª  — t          | ¦  «        }|| _        |�|| _        |dk    rt          j        | j        |fi |¤Žnt          j        ¦   «         | _        | j                             | j	        ¦  «         | j
        �W|dk    rt          j        | j        |fi |¤Žnt          j        ¦   «         | _
        | j
                             | j	        ¦  «         dS dS )z›Reset classifier head.

        Args:
            num_classes: Number of classes for new classifier.
            global_pool: Global pooling type.
        Nr   )r   rð   rñ   r3   r4   r
  rr   r  ÚtrainrŠ   r  )r=   rð   rñ   r>   s       rA   Úreset_classifierzVOLO.reset_classifier«  só   € õ ˜dÑ#Ô#ˆØ&ˆÔØÐ"Ø*ˆDÔà5@À1²_°_õ ”IØÔ˜{ð2ð 2Ø.0ð2ð 2ð 2ÝJLÌ+É-Ì-ð 	Œ	àŒ	�Š˜œÑ&Ô&Ð&ØŒ=Ð$à9DÀqº¸õ œIØÔ! ;ð6ð 6Ø24ð6ð 6ð 6ÝNPÌkÉmÌmð ŒMàŒM×Ò ¤Ñ.Ô.Ð.Ð.Ð.ð %Ð$rB   rC   c                 óL  — t          | j        ¦  «        D ]k\  }}|dk    r|| j        z   }|                      |¦  «        }| j        r/t
          j                             ¦   «         st          ||¦  «        }Œ` ||¦  «        }Œl|j	        \  }}}}| 
                    |d|¦  «        }|S )z·Forward pass through token processing stages.

        Args:
            x: Input tensor of shape (B, H, W, C).

        Returns:
            Token tensor of shape (B, N, C).
        rG   rH   )Ú	enumerater  r  r  r  ra   r�   Úis_scriptingr   rK   rO   )r=   rC   ÚidxÚblockrS   rT   rU   rV   s           rA   Úforward_tokenszVOLO.forward_tokens¾  s°   € õ $ D¤LÑ1Ô1ð 	ð 	‰JˆC�Ø�aŠxˆxà˜œÑ&�Ø—M’M !Ñ$Ô$�ØÔ&ð ­u¬y×/EÒ/EÑ/GÔ/Gð Ý˜u aÑ(Ô(��à�E˜!‘H”H��à”W‰
ˆˆ1ˆa�Ø�IŠI�a˜˜QÑÔˆØˆrB   c                 ó  — |j         \  }}}| j                             |dd¦  «        }t          j        ||gd¬¦  «        }| j        D ]C}| j        r/t          j                             ¦   «         st          ||¦  «        }Œ8 ||¦  «        }ŒD|S )zÍForward pass through class attention blocks.

        Args:
            x: Input token tensor of shape (B, N, C).

        Returns:
            Output tensor with class token of shape (B, N+1, C).
        rH   r   rI   )
rK   r  Úexpandra   r£   r  r  r�   r=  r   )r=   rC   rS   rœ   rV   Ú
cls_tokensr?  s          rA   Úforward_clszVOLO.forward_clsÕ  s™   € ð ”'‰ˆˆ1ˆaØ”^×*Ò*¨1¨b°"Ñ5Ô5ˆ
ÝŒI�z 1�o¨1Ð-Ñ-Ô-ˆØÔ&ð 	ð 	ˆEØÔ&ð ­u¬y×/EÒ/EÑ/GÔ/Gð Ý˜u aÑ(Ô(��à�E˜!‘H”H��ØˆrB   c                 óÈ  — 	 |                       |¦  «        }|                     dddd¦  «        }| j        �r| j        �r	t          j                             | j        | j        ¦  «                             ¦   «         }|j	        d         | j
        z  |j	        d         | j
        z  }}t          |                     ¦   «         || j
        ¬¦  «        \  }}}}|                     ¦   «         }	| j
        |z  | j
        |z  }}
| j
        |z  | j
        |z  }}|                     d¦  «        dd…|
|…||…dd…f         |	dd…|
|…||…dd…f<   |	}nd\  }}}}|                      |¦  «        }| j        �|                      |¦  «        }|                      |¦  «        }| j        dk    r|                     d¬	¦  «        }n| j        d
k    r|dd…df         }n|}| j        €|S |                      |dd…dd…f         ¦  «        }| j        s!|d|                     d¦  «        d         z  z   S | j        r±| j        rª|                     |j	        d         |||j	        d         ¦  «        }|                     ¦   «         }	|                     d¦  «        dd…||…||…dd…f         |	dd…||…||…dd…f<   |	}|                     |j	        d         ||z  |j	        d         ¦  «        }||||||ffS )a  Forward pass for training with mix token support.

        Args:
            x: Input tensor of shape (B, C, H, W).

        Returns:
            If training with mix_token: tuple of (class_token, aux_tokens, bbox).
            Otherwise: class_token tensor.
        r   rG   r   r   )r2   N)r   r   r   r   ÚavgrI   rï   ç      à?rH   )r  rL   r	  rŠ   ra   ÚdistributionsÚBetar  ÚsamplerK   rü   r¿   r©   ÚcloneÚflipr@  r  rD  r  rñ   Úmeanr  ÚmaxrO   )r=   rC   rª   Úpatch_hÚpatch_wr»   r¼   r½   r¾   Útemp_xÚsbbx1Úsbby1Úsbbx2Úsbby2Úx_clsÚx_auxs                   rA   Úforward_trainzVOLO.forward_trainè  s  € ð	ð ×Ò˜QÑÔˆØ�IŠI�a˜˜A˜qÑ!Ô!ˆð Œ>ñ 
	0˜dœmñ 
	0ÝÔ%×*Ò*¨4¬9°d´iÑ@Ô@×GÒGÑIÔIˆCØ œw qœz¨TÔ-?Ñ?ÀÄÈÄÈtÔOaÑAa�WˆGÝ%.¨q¯vªv©x¬x¸ÀDÔDVÐ%WÑ%WÔ%WÑ"ˆD�$˜˜dØ—W’W‘Y”YˆFØÔ-°Ñ4°dÔ6HÈ4Ñ6O�5ˆEØÔ-°Ñ4°dÔ6HÈ4Ñ6O�5ˆEØ56·V²V¸A±Y´Y¸q¸q¸qÀ%ÈÀ+ÈuÐUZÈ{Ð\]Ð\]Ð\]Ð?]Ô5^ˆF�1�1�1�e˜E�k 5¨ ;°°°Ð1Ñ2ØˆAˆAà%/Ñ"ˆD�$˜˜dð ×Ò Ñ"Ô"ˆð ÔÐ(Ø× Ò  Ñ#Ô#ˆAØ�IŠI�a‰LŒLˆàÔ˜uÒ$Ð$Ø—F’F˜q�F‘M”MˆEˆEØÔ Ò(Ð(Ø�a�a�a˜�d”GˆEˆEàˆEàŒ=Ð ØˆLà—’˜a    1 2 2 œhÑ'Ô'ˆØŒ}ð 	1Ø˜3 §¢¨1¡¤¨a¤Ñ0Ñ0Ð0àŒ>ð 	V˜dœmð 	VØ—M’M %¤+¨a¤.°'¸7ÀEÄKÐPRÄOÑTÔTˆEØ—[’[‘]”]ˆFØ16·²¸A±´¸q¸q¸qÀ$ÀtÀ)ÈTÐRVÈYÐXYÐXYÐXYÐ?YÔ1ZˆF�1�1�1�d˜4�i  d ¨A¨A¨AÐ-Ñ.ØˆEØ—M’M %¤+¨a¤.°'¸GÑ2CÀUÄ[ÐQSÄ_ÑUÔUˆEð �e˜d D¨$°Ð5Ð5Ð5rB   ÚNCHWÚindicesr  Ú
stop_earlyÚ
output_fmtÚintermediates_onlyc           	      óÒ  ‡ — |dv s
J d¦   «         ‚g }t          t          ‰ j        ¦  «        |¦  «        \  }}	ˆ fd„|D ¦   «         }‰ j        |	         }	|j        \  }
}}}‰                      |¦  «                             dddd¦  «        }t          j                             ¦   «         s|s‰ j	        }n‰ j	        d|	dz   …         }t          |¦  «        D ]º\  }}|dk    r|‰ j        z   }‰                      |¦  «        }‰ j        r/t          j                             ¦   «         st          ||¦  «        }n ||¦  «        }||v rK|r|dk    r‰                      |¦  «        }n|}|                     |                     dddd¦  «        ¦  «         Œ»|r|S |j        \  }
}}}|                     |
d	|¦  «        }‰ j        �‰                      |¦  «        }‰                      |¦  «        }||fS )
aè   Forward features that returns intermediates.

        Args:
            x: Input image tensor
            indices: Take last n blocks if int, all if None, select matching indices if sequence
            norm: Apply norm layer to all intermediates
            stop_early: Stop iterating over blocks when last desired intermediate hit
            output_fmt: Shape of intermediate feature outputs
            intermediates_only: Only return intermediate features
        Returns:

        )rY  zOutput format must be NCHW.c                 ó*   •— g | ]}‰j         |         ‘ŒS rp   )r  )r  r  r=   s     €rA   r  z.VOLO.forward_intermediates.<locals>.<listcomp>=  s    ø€ ÐAÐAÐA¨q˜œ¨Ô*ÐAÐAÐArB   r   rG   r   r   NrH   )r   r  r  rK   r  rL   ra   r�   r=  r  r<  r  r  r  r   r  ræ   rO   r  rD  )r=   rC   rZ  r  r[  r\  r]  ÚintermediatesÚtake_indicesÚ	max_indexrS   Ú_ÚheightÚwidthr  r>  r?  Úx_interrT   rU   rV   s   `                    rA   Úforward_intermediateszVOLO.forward_intermediates%  s  ø€ ð* ˜YÐ&Ð&Ð&Ð(EÑ&Ô&Ð&ØˆÝ"6µs¸4¼?Ñ7KÔ7KÈWÑ"UÔ"UÑˆ�iØAÐAÐAÐA°LÐAÑAÔAˆØ”O IÔ.ˆ	ð  œgÑˆˆ1ˆf�eØ×Ò˜QÑÔ×'Ò'¨¨1¨a°Ñ3Ô3ˆõ Œ9×!Ò!Ñ#Ô#ð 	3¨:ð 	3Ø”lˆGˆGà”l > I°¡M >Ô2ˆGÝ# GÑ,Ô,ð 	Bð 	B‰JˆC�Ø�aŠxˆxà˜œÑ&�Ø—M’M !Ñ$Ô$�ØÔ&ð ­u¬y×/EÒ/EÑ/GÔ/Gð Ý˜u aÑ(Ô(��à�E˜!‘H”H�Ø�lÐ"Ð"Øð  ˜C 1šH˜HØ"Ÿiši¨™lœl�G�Gà�GØ×$Ò$ W§_¢_°Q¸¸1¸aÑ%@Ô%@ÑAÔAÐAøàð 	!Ø Ð ð ”W‰
ˆˆ1ˆa�Ø�IŠI�a˜˜QÑÔˆØÔÐ(Ø× Ò  Ñ#Ô#ˆAØ�IŠI�a‰LŒLˆà�-ÐÐrB   r   Ú
prune_normÚ
prune_headc                 ó.  — 	 t          t          | j        ¦  «        |¦  «        \  }}| j        |         }| j        d|dz   …         | _        |rt	          j        ¦   «         | _        |r.t	          j        ¦   «         | _        |  	                    dd¦  «         |S )aH  Prune layers not required for specified intermediates.

        Args:
            indices: Indices of intermediate layers to keep.
            prune_norm: Whether to prune normalization layer.
            prune_head: Whether to prune classification head.

        Returns:
            List of kept intermediate indices.
        Nr   r   Ú )
r   r  r  r  r3   rr   r  r  r  r:  )r=   rZ  rh  ri  ra  rb  s         rA   Úprune_intermediate_layerszVOLO.prune_intermediate_layersf  sŽ   € ð 	å"6µs¸4¼?Ñ7KÔ7KÈWÑ"UÔ"UÑˆ�iØ”O IÔ.ˆ	Ø”| N Y°¡] NÔ3ˆŒØð 	&Ýœ™œˆDŒIØð 	)Ý "¤¡¤ˆDÔØ×!Ò! ! RÑ(Ô(Ð(ØÐrB   c                 óè   — |                       |¦  «                             dddd¦  «        }|                      |¦  «        }| j        �|                      |¦  «        }|                      |¦  «        }|S )z¡Forward pass through feature extraction.

        Args:
            x: Input tensor of shape (B, C, H, W).

        Returns:
            Feature tensor.
        r   rG   r   r   )r  rL   r@  r  rD  r  rz   s     rA   Úforward_featureszVOLO.forward_features‚  sr   € ð ×Ò˜QÑÔ×'Ò'¨¨1¨a°Ñ3Ô3ˆð ×Ò Ñ"Ô"ˆð ÔÐ(Ø× Ò  Ñ#Ô#ˆAØ�IŠI�a‰LŒLˆØˆrB   Ú
pre_logitsc                 ól  — | j         dk    r|                     d¬¦  «        }n| j         dk    r|dd…df         }n|}|                      |¦  «        }|r|S |                      |¦  «        }| j        �B|                      |dd…dd…f         ¦  «        }|d|                     d¦  «        d         z  z   }|S )zñForward pass through classification head.

        Args:
            x: Input feature tensor.
            pre_logits: Whether to return pre-logits features.

        Returns:
            Classification logits or pre-logits features.
        rF  r   rI   rï   Nr   rG  )rñ   rM  r  r  r  rN  )r=   rC   ro  ÚoutÚauxs        rA   Úforward_headzVOLO.forward_head–  sÃ   € ð Ô˜uÒ$Ð$Ø—&’&˜Q�&‘-”-ˆCˆCØÔ Ò(Ð(Ø�A�A�A�q�D”'ˆCˆCàˆCØ�NŠN˜1ÑÔˆØð 	ØˆJØ�iŠi˜‰nŒnˆØŒ=Ð$à—-’-  ! ! ! Q R R %¤Ñ)Ô)ˆCØ˜˜cŸgšg a™jœj¨œmÑ+Ñ+ˆCØˆ
rB   c                 ó\   — 	 |                       |¦  «        }|                      |¦  «        }|S )z¶Forward pass (simplified, without mix token training).

        Args:
            x: Input tensor of shape (B, C, H, W).

        Returns:
            Classification logits.
        )rn  rs  rz   s     rA   rY   zVOLO.forward°  s3   € ð 	@Ø×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆrB   ©F)T)N)NFFrY  F)r   FT)'rZ   r[   r\   r]   r3   r}   r   r^   Ústrr   r   r_   r`   r
   r~   r1   r  ra   r�   ÚignoreÚsetr,  r   r   r2  r5  r7  r:  rb   r@  rD  r	   rX  rg  rl  rn  rs  rY   rc   rd   s   @rA   r   r   Ä  s  ø€ € € € € Ø(Ð(ð
  ØØ#Ø&ØØ#%Ø.2Ø-1Ø,GØ2MØ"Ø"Ø!Ø#%Ø$&Ø$&Ø*,¬,Ø5AØ!%Ø"'Ø!"ØØð3e'ð e'à˜”Iðe'ð ðe'ð ð	e'ð
 ðe'ð ðe'ð ðe'ð !ðe'ð !  c¤Ô+ðe'ð    S¤	Ô*ðe'ð ˜t S˜yÔ)ðe'ð  % T¨3 YÔ/ðe'ð ðe'ð ðe'ð ðe'ð  !ð!e'ð" "ð#e'ð$ "ð%e'ð& ˜RœYœð'e'ð( " %¨¨S¨¤/Ô2ð)e'ð* ð+e'ð,  ð-e'ð. ð/e'ð e'ð e'ð e'ð e'ð e'ðN	-˜rœyð 	-¨Tð 	-ð 	-ð 	-ð 	-ð „YÔð* ð *ð *ð *ñ Ôð*ð „YÔð
ð 
 Dð 
°T¸#¸s¸(´^ð 
ð 
ð 
ñ Ôð
ð, „YÔð)ð )¨Tð )¸Tð )ð )ð )ñ Ôð)ð „YÔð ¤	ð ð ð ñ Ôðð/ð /¨Cð /¸hÀs¼mð /ÐW[ð /ð /ð /ð /ð& ¤ð °´ð ð ð ð ð.˜Uœ\ð ¨e¬lð ð ð ð ð&;6˜uœ|ð ;6°°e´lÀEÈ%Ì,ÐX]ÔXdÐfkÐloÐqtÐvyÐ{~Ðl~ÔfÐJô  EAð  7Aô  1Bð ;6ð ;6ð ;6ð ;6ð@ 8<ØØ$Ø$Ø',ð? ð ? àŒ|ð? ð ˜e C¨¨c¬ NÔ3Ô4ð? ð ð	? ð
 ð? ð ð? ð !%ð? ð 
ˆt�E”LÔ! 5¨¬°t¸E¼LÔ7IÐ)IÔ#JÐJÔ	Kð? ð ? ð ? ð ? ðF ./Ø$Ø#ð	ð à˜3  S¤	˜>Ô*ðð ðð ð	ð
 
ˆcŒðð ð ð ð8 %¤,ð °5´<ð ð ð ð ð(ð ˜eœlð ¸ð ÈÌð ð ð ð ð4˜œð ¨%¬,ð ð ð ð ð ð ð ð rB   ÚvariantÚ
pretrainedc                 óx   — |                      dd¦  «        }t          t          | |fdt          |d¬¦  «        i|¤ŽS )zØCreate VOLO model.

    Args:
        variant: Model variant name.
        pretrained: Whether to load pretrained weights.
        **kwargs: Additional model arguments.

    Returns:
        VOLO model instance.
    Úout_indicesr   Úfeature_cfgÚgetter)r|  Úfeature_cls)Úpopr   r   r  )ry  rz  r¥   r|  s       rA   Ú_create_volor�  ¿  sX   € ð —*’*˜]¨AÑ.Ô.€KÝÝØØðð õ  [¸hÐGÑGÔGð	ð
 ðð ð rB   rk  Úurlc                 ó8   — | ddddddt           t          ddd	d
œ|¥S )z¾Create model configuration.

    Args:
        url: URL for pretrained weights.
        **kwargs: Additional configuration options.

    Returns:
        Model configuration dictionary.
    rî   )r   rÂ   rÂ   Nç¸…ëQ¸î?ÚbicubicTzpatch_embed.conv.0)r  r  z
apache-2.0)r‚  rð   Ú
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚfixed_input_sizerM  r  Ú
first_convÚ
classifierÚlicenser   )r‚  r¥   s     rA   Ú_cfgrŽ  Ô  s=   € ð Ø¨=ÀtØ¨)ÈÝ%Õ.BØ*Ð:NØðð ð ðð rB   ztimm/zLhttps://github.com/sail-sg/volo/releases/download/volo_1/d1_224_84.2.pth.tarr„  )Ú	hf_hub_idr‚  rˆ  zLhttps://github.com/sail-sg/volo/releases/download/volo_1/d1_384_85.2.pth.tarr¬   )r   rÄ   rÄ   )r�  r‚  rˆ  r†  zLhttps://github.com/sail-sg/volo/releases/download/volo_1/d2_224_85.2.pth.tarzLhttps://github.com/sail-sg/volo/releases/download/volo_1/d2_384_86.0.pth.tarzLhttps://github.com/sail-sg/volo/releases/download/volo_1/d3_224_85.4.pth.tarzLhttps://github.com/sail-sg/volo/releases/download/volo_1/d3_448_86.3.pth.tar)r   éÀ  r�  zLhttps://github.com/sail-sg/volo/releases/download/volo_1/d4_224_85.7.pth.tarzMhttps://github.com/sail-sg/volo/releases/download/volo_1/d4_448_86.79.pth.targffffffò?zMhttps://github.com/sail-sg/volo/releases/download/volo_1/d5_224_86.10.pth.tarzLhttps://github.com/sail-sg/volo/releases/download/volo_1/d5_448_87.0.pth.tarzMhttps://github.com/sail-sg/volo/releases/download/volo_1/d5_512_87.07.pth.tar)r   é   r‘  )zvolo_d1_224.sail_in1kzvolo_d1_384.sail_in1kzvolo_d2_224.sail_in1kzvolo_d2_384.sail_in1kzvolo_d3_224.sail_in1kzvolo_d3_448.sail_in1kzvolo_d4_224.sail_in1kzvolo_d4_448.sail_in1kzvolo_d5_224.sail_in1kzvolo_d5_448.sail_in1kzvolo_d5_512.sail_in1kc                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	úVOLO-D1 model, Params: 27M.©r.   r.   r‚   rG   ©éÀ   rÄ   rÄ   rÄ   ©é   é   r™  r™  ©rà   ró   r!   Úvolo_d1_224rz  rp   )r›  ©r  r�  ©rz  r¥   Ú
model_argsÚmodels       rA   r›  r›    ó@   € õ Ðp˜\Ð6JÐVeÐpÐpÐioÐpÐp€JÝÐLÐL°:ÐLÀÐLÐL€EØ€LrB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	r“  r”  r•  r—  rš  Úvolo_d1_384rz  rp   )r¢  rœ  r�  s       rA   r¢  r¢  !  r   rB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	úVOLO-D2 model, Params: 59M.©r˜  r.   é
   r.   ©é   r‘  r‘  r‘  ©r‚   rÍ   rÍ   rÍ   rš  Úvolo_d2_224rz  rp   )rª  rœ  r�  s       rA   rª  rª  )  ó@   € õ Ðq˜]Ð7KÐWfÐqÐqÐjpÐqÐq€JÝÐLÐL°:ÐLÀÐLÐL€EØ€LrB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	r¤  r¥  r§  r©  rš  Úvolo_d2_384rz  rp   )r­  rœ  r�  s       rA   r­  r­  1  r«  rB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	úVOLO-D3 model, Params: 86M.©r‚   r‚   rÍ   r.   r§  r©  rš  Úvolo_d3_224rz  rp   )r±  rœ  r�  s       rA   r±  r±  9  r«  rB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	r¯  r°  r§  r©  rš  Úvolo_d3_448rz  rp   )r³  rœ  r�  s       rA   r³  r³  A  r«  rB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	úVOLO-D4 model, Params: 193M.r°  ©rÄ   é   r·  r·  ©r™  rÍ   rÍ   rÍ   rš  Úvolo_d4_224rz  rp   )r¹  rœ  r�  s       rA   r¹  r¹  I  ó@   € õ Ðr˜]Ð7KÐWgÐrÐrÐkqÐrÐr€JÝÐLÐL°:ÐLÀÐLÐL€EØ€LrB   c                 óB   — t          dddddœ|¤Ž}t          dd| i|¤Ž}|S )	rµ  r°  r¶  r¸  rš  Úvolo_d4_448rz  rp   )r¼  rœ  r�  s       rA   r¼  r¼  Q  rº  rB   c           	      óF   — t          d	ddddddœ|¤Ž}t          d
d| i|¤Ž}|S )újVOLO-D5 model, Params: 296M.

    stem_hidden_dim=128, the dim in patch embedding is 128 for VOLO-D5.
    ©r™  r™  é   r.   r¶  r¸  r.   é€   ©rà   ró   r!   rh   rò   Úvolo_d5_224rz  rp   )rÃ  rœ  r�  s       rA   rÃ  rÃ  Y  óR   € õ ð 4ØÐ+?ÐK[Ø Sð4ð 4à,2ð4ð 4€Jõ ÐLÐL°:ÐLÀÐLÐL€EØ€LrB   c           	      óF   — t          d	ddddddœ|¤Ž}t          d
d| i|¤Ž}|S )r¾  r¿  r¶  r¸  r.   rÁ  rÂ  Úvolo_d5_448rz  rp   )rÆ  rœ  r�  s       rA   rÆ  rÆ  f  rÄ  rB   c           	      óF   — t          d	ddddddœ|¤Ž}t          d
d| i|¤Ž}|S )r¾  r¿  r¶  r¸  r.   rÁ  rÂ  Úvolo_d5_512rz  rp   )rÈ  rœ  r�  s       rA   rÈ  rÈ  s  rÄ  rB   r­   )
r   r   r   rG   rg   Fr   r   NN)rg   Fr   r   ru  )rk  )Gr]   rM   Útypingr   r   r   r   r   r   r	   r
   ra   Útorch.nnr3   Útorch.nn.functionalÚ
functionalrQ   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   Ú_builderr   Ú	_featuresr   Ú_manipulater   Ú	_registryr   r   Ú__all__r~   r   rf   r€   r“   r˜   rŸ   rv  r¨   r^   r`   r¿   rÁ   rÙ   r_   rÑ   rê   rì   r   r�  rŽ  Údefault_cfgsr›  r¢  rª  r­  r±  r³  r¹  r¼  rÃ  rÆ  rÈ  rp   rB   rA   ú<module>rÕ     s?	  ððð ð* €€€Ø JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ Jà €€€Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ AØ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lØ *Ð *Ð *Ð *Ð *Ð *Ø +Ð +Ð +Ð +Ð +Ð +Ø #Ð #Ð #Ð #Ð #Ð #Ø <Ð <Ð <Ð <Ð <Ð <Ð <Ð <àˆ(€ðOð Oð Oð Oð O�r”yñ Oô Oð OðdEð Eð Eð Eð E�”	ñ Eô Eð EðPAð Að Að Að A�”	ñ Aô Að AðH1ð 1ð 1ð 1ð 1�"”)ñ 1ô 1ð 1ðh>ð >ð >ð >ð >�R”Yñ >ô >ð >ðBD7ð D7ð D7ð D7ð D7�”ñ D7ô D7ð D7ðN:˜#ð :¨ð :°´ð :ð :ð :ð :ð >ð >�E˜#˜s˜(”Oð >¨%ð >¸ð >ÀEÈ#ÈsÐTWÐY\ÐJ\ÔD]ð >ð >ð >ð >ðBCð Cð Cð Cð C�”ñ Cô Cð CðL"ð "ð "ð "ð "�”ñ "ô "ð "ðT ØØØØØØØ "ØØð7ð 7Øð7àð7ð ð7ð �S”	ð	7ð
 ð7ð ð7ð ð7ð ð7ð ð7ð ð7ð ð7ð ð7ð ð7ð  „]ð!7ð 7ð 7ð 7ð@ ØØØ "ð*ð *Øð*àð*ð ð*ð �S”	ð	*ð
 ð*ð ð*ð ð*ð ð*ð ð*ð ð*ð „]ð*ð *ð *ð *ðZxð xð xð xð xˆ2Œ9ñ xô xð xðvð ˜#ð ¨4ð À3ð È4ð ð ð ð ð*ð ˆcð  #ð ¨$¨s°C¨x¬.ð ð ð ð ð* %Ð$Ø!˜TØØZØðñ ô ð "˜TØØZØ ð0ñ 0ô 0ð "˜TØØZØðñ ô ð "˜TØØZØ ð0ñ 0ô 0ð "˜TØØZØðñ ô ð "˜TØØZØ ð0ñ 0ô 0ð "˜TØØZØðñ ô ð "˜TØØ[Ø -ð1ñ 1ô 1ð "˜TØØ[Øðñ ô ð "˜TØØZØ -ð1ñ 1ô 1ð "˜TØØ[Ø -ð1ñ 1ô 1ðS-&ð -&ñ -ô -€ð` ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ðð ˜Dð °Cð ¸Dð ð ð ñ „ðð ð	ð 	˜Dð 	°Cð 	¸Dð 	ð 	ð 	ñ „ð	ð ð	ð 	˜Dð 	°Cð 	¸Dð 	ð 	ð 	ñ „ð	ð ð	ð 	˜Dð 	°Cð 	¸Dð 	ð 	ð 	ñ „ð	ð 	ð 	rB   