§
    ·¯¸j+&  ã            	       óî  — d Z ddlZddlmZmZ ddlmZ ddlmZm	Z	 ddl
mZmZmZ ddlmZ dd	lmZmZ dd
lmZ  G d„ dej        ¦  «        Z G d„ dej        ¦  «        Zdd„Zdd„Z e ed¬¦  «         ed¬¦  «         e¦   «          ed¬¦  «         ed¬¦  «        dœ¦  «        Zeddefd„¦   «         Zeddefd„¦   «         Zeddefd„¦   «         Zeddefd„¦   «         Zeddefd„¦   «         ZdS ) aß   Selective Kernel Networks (ResNet base)

Paper: Selective Kernel Networks (https://arxiv.org/abs/1903.06586)

This was inspired by reading 'Compounding the Performance Improvements...' (https://arxiv.org/abs/2001.06268)
and a streamlined impl at https://github.com/clovaai/assembled-cnn but I ended up building something closer
to the original paper with some modifications of my own to better balance param count vs accuracy.

Hacked together by / Copyright 2020 Ross Wightman
é    N)ÚOptionalÚType)Únn©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚSelectiveKernelÚConvNormActÚcreate_attné   )Úbuild_model_with_cfg)Úregister_modelÚgenerate_default_cfgs)ÚResNetc            "       óp  ‡ — e Zd ZdZddddddddej        ej        ddddddfdedededeej	                 ded	ed
ee
         dededee         deej	                 deej	                 deeej	                          deeej	                          deej	                 deej	                 f ˆ fd„Zd„ Zd„ Zˆ xZS )ÚSelectiveKernelBasicr   Né@   ÚinplanesÚplanesÚstrideÚ
downsampleÚcardinalityÚ
base_widthÚ	sk_kwargsÚreduce_firstÚdilationÚfirst_dilationÚ	act_layerÚ
norm_layerÚ
attn_layerÚaa_layerÚ
drop_blockÚ	drop_pathc                 ó¤  •— ||dœ}t          ¦   «                              ¦   «          |pi }t          d||dœ|¤Ž}|dk    s
J d¦   «         ‚|dk    s
J d¦   «         ‚||z  }|| j        z  }|
p|	}
t	          ||f||
||dœ|¤|¤Ž| _        t          ||fd|	d	d
œ|¤Ž| _        t          ||fi |¤Ž| _	         |d¬¦  «        | _
        || _        || _        d S )N©ÚdeviceÚdtype©r   r   r   z)BasicBlock only supports cardinality of 1r   z0BasicBlock doest not support changing base width)r   r   r!   Ú
drop_layeré   F)Úkernel_sizer   Ú	apply_actT©Úinplace© )ÚsuperÚ__init__ÚdictÚ	expansionr	   Úconv1r
   Úconv2r   ÚseÚactr   r#   )Úselfr   r   r   r   r   r   r   r   r   r   r   r   r    r!   r"   r#   r&   r'   ÚddÚconv_kwargsÚfirst_planesÚ	outplanesÚ	__class__s                          €úa/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/sknet.pyr1   zSelectiveKernelBasic.__init__   s[  ø€ ð* ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐà�O ˆ	ÝÐL Y¸:ÐLÐLÈÐLÐLˆØ˜aÒÐÐÐ!LÑÔÐØ˜RÒÐÐÐ!SÑÔÐØ Ñ-ˆØ˜Tœ^Ñ+ˆ	Ø'Ð3¨8ˆå$ØØð	
ð Ø#ØØ!ð	
ð 	
ð ð	
ð ð	
ð 	
ˆŒ
õ !ØØð
ð ØØð
ð 
ð ð
ð 
ˆŒ
õ ˜j¨)Ð:Ð:°rÐ:Ð:ˆŒØ�9 TÐ*Ñ*Ô*ˆŒØ$ˆŒØ"ˆŒˆˆó    c                 óœ   — t          | j        j        dd ¦  «        �0t          j                             | j        j        j        ¦  «         d S d S ©NÚweight)Úgetattrr5   Úbnr   ÚinitÚzeros_rB   ©r8   s    r>   Úzero_init_lastz#SelectiveKernelBasic.zero_init_lastQ   óA   € Ý�4”:”= (¨DÑ1Ô1Ð=ÝŒG�NŠN˜4œ:œ=Ô/Ñ0Ô0Ð0Ð0Ð0ð >Ð=r?   c                 ó:  — |}|                       |¦  «        }|                      |¦  «        }| j        �|                      |¦  «        }| j        �|                      |¦  «        }| j        �|                      |¦  «        }||z  }|                      |¦  «        }|S ©N)r4   r5   r6   r#   r   r7   ©r8   ÚxÚshortcuts      r>   ÚforwardzSelectiveKernelBasic.forwardU   s‹   € ØˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØŒ7ÐØ—’˜‘
”
ˆAØŒ>Ð%Ø—’˜qÑ!Ô!ˆAØŒ?Ð&Ø—’ xÑ0Ô0ˆHØ	ˆX‰ˆØ�HŠH�Q‰KŒKˆØˆr?   ©Ú__name__Ú
__module__Ú__qualname__r3   r   ÚReLUÚBatchNorm2dÚintr   ÚModuler2   r   r1   rH   rO   Ú__classcell__©r=   s   @r>   r   r      sy  ø€ € € € € Ø€Ið Ø.2Ø Ø Ø(,Ø !ØØ,0Ø)+¬Ø*,¬.Ø48Ø26Ø.2Ø-1ØØð'5#ð 5#àð5#ð ð5#ð ð	5#ð
 ! ¤Ô+ð5#ð ð5#ð ð5#ð   ”~ð5#ð ð5#ð ð5#ð % SœMð5#ð ˜BœI”ð5#ð ˜RœYœð5#ð !  b¤i¤Ô1ð5#ð ˜t B¤IœÔ/ð5#ð  ! ¤Ô+ð!5#ð"   ¤	Ô*ð#5#ð 5#ð 5#ð 5#ð 5#ð 5#ðn1ð 1ð 1ðð ð ð ð ð ð r?   r   c            "       óp  ‡ — e Zd ZdZddddddddej        ej        ddddddfdedededeej	                 d	ed
edee
         dededee         deej	                 deej	                 deeej	                          deeej	                          deej	                 deej	                 f ˆ fd„Zd„ Zd„ Zˆ xZS )ÚSelectiveKernelBottlenecké   r   Nr   r   r   r   r   r   r   r   r   r   r   r   r   r    r!   r"   r#   c           	      óâ  •— ||dœ}t          ¦   «                              ¦   «          |pi }t          d||dœ|¤Ž}t          t	          j        ||dz  z  ¦  «        |z  ¦  «        }||z  }|| j        z  }|
p|	}
t          ||fddi|¤Ž| _        t          ||f||
|||dœ|¤|¤Ž| _
        t          ||fdddœ|¤Ž| _        t          ||fi |¤Ž| _         |d	¬
¦  «        | _        || _        || _        d S )Nr%   r(   r   r+   r   )r   r   Úgroupsr!   r)   F)r+   r,   Tr-   r/   )r0   r1   r2   rV   ÚmathÚfloorr3   r
   r4   r	   r5   Úconv3r   r6   r7   r   r#   )r8   r   r   r   r   r   r   r   r   r   r   r   r   r    r!   r"   r#   r&   r'   r9   r:   Úwidthr;   r<   r=   s                           €r>   r1   z"SelectiveKernelBottleneck.__init__g   sV  ø€ ð* ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐà�O ˆ	ÝÐL Y¸:ÐLÐLÈÐLÐLˆÝ•D”J˜v¨°b©Ñ9Ñ:Ô:¸[ÑHÑIÔIˆØ Ñ,ˆØ˜Tœ^Ñ+ˆ	Ø'Ð3¨8ˆå  ¨<ÐVÐVÀQÐVÈ+ÐVÐVˆŒ
Ý$ØØð

ð Ø#ØØØ!ð

ð 

ð ð

ð ð

ð 

ˆŒ
õ ! ¨	Ða¸qÈEÐaÐaÐU`ÐaÐaˆŒ
Ý˜j¨)Ð:Ð:°rÐ:Ð:ˆŒØ�9 TÐ*Ñ*Ô*ˆŒØ$ˆŒØ"ˆŒˆˆr?   c                 óœ   — t          | j        j        dd ¦  «        �0t          j                             | j        j        j        ¦  «         d S d S rA   )rC   ra   rD   r   rE   rF   rB   rG   s    r>   rH   z(SelectiveKernelBottleneck.zero_init_last˜   rI   r?   c                 ód  — |}|                       |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        �|                      |¦  «        }| j        �|                      |¦  «        }| j        �|                      |¦  «        }||z  }|                      |¦  «        }|S rK   )r4   r5   ra   r6   r#   r   r7   rL   s      r>   rO   z!SelectiveKernelBottleneck.forwardœ   sš   € ØˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØŒ7ÐØ—’˜‘
”
ˆAØŒ>Ð%Ø—’˜qÑ!Ô!ˆAØŒ?Ð&Ø—’ xÑ0Ô0ˆHØ	ˆX‰ˆØ�HŠH�Q‰KŒKˆØˆr?   rP   rY   s   @r>   r[   r[   d   sy  ø€ € € € € Ø€Ið Ø.2Ø Ø Ø(,Ø !ØØ,0Ø)+¬Ø*,¬.Ø48Ø26Ø.2Ø-1ØØð'/#ð /#àð/#ð ð/#ð ð	/#ð
 ! ¤Ô+ð/#ð ð/#ð ð/#ð   ”~ð/#ð ð/#ð ð/#ð % SœMð/#ð ˜BœI”ð/#ð ˜RœYœð/#ð !  b¤i¤Ô1ð/#ð ˜t B¤IœÔ/ð/#ð  ! ¤Ô+ð!/#ð"   ¤	Ô*ð#/#ð /#ð /#ð /#ð /#ð /#ðb1ð 1ð 1ðð ð ð ð ð ð r?   r[   Fc                 ó*   — t          t          | |fi |¤ŽS rK   )r   r   )ÚvariantÚ
pretrainedÚkwargss      r>   Ú_create_skresnetri   ¬   s,   € ÝÝØØðð ð ð	ð ð r?   Ú c                 ó6   — | dddddt           t          dddd	œ|¥S )
Niè  )r*   éà   rl   )é   rm   g      ì?Úbicubicr4   Úfcz
apache-2.0)ÚurlÚnum_classesÚ
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚmeanÚstdÚ
first_convÚ
classifierÚlicenser   )rp   rh   s     r>   Ú_cfgr{   µ   s8   € àØ¨=ÀvØ¨IÝ%Õ.BØ¨TØðð ð ðð r?   ztimm/)Ú	hf_hub_idzconv1.0)rx   )zskresnet18.ra_in1kzskresnet34.ra_in1kzskresnet50.untrainedzskresnet50d.untrainedzskresnext50_32x4d.ra_in1kÚreturnc           	      ó¨   — t          ddd¬¦  «        }t          t          g d¢t          |¬¦  «        d¬¦  «        }t          d	| fi t          |fi |¤Ž¤ŽS )
züConstructs a Selective Kernel ResNet-18 model.

    Different from configs in Select Kernel paper or "Compounding the Performance Improvements..." this
    variation splits the input channels to the selective convolutions to keep param count down.
    ç      À?é   T©Úrd_ratioÚ
rd_divisorÚsplit_input)é   r…   r…   r…   ©r   F©ÚblockÚlayersÚ
block_argsrH   Ú
skresnet18©r2   r   ri   ©rg   rh   r   Ú
model_argss       r>   r‹   r‹   Ë   óp   € õ ˜e°ÀÐEÑEÔE€IÝÝ"¨<¨<¨<ÅDÐS\ÐD]ÑD]ÔD]Øðñ ô €Jõ ˜L¨*ÐSÐS½¸ZÐ8RÐ8RÈ6Ð8RÐ8RÐSÐSÐSr?   c           	      ó¨   — t          ddd¬¦  «        }t          t          g d¢t          |¬¦  «        d¬¦  «        }t          d	| fi t          |fi |¤Ž¤ŽS )
züConstructs a Selective Kernel ResNet-34 model.

    Different from configs in Select Kernel paper or "Compounding the Performance Improvements..." this
    variation splits the input channels to the selective convolutions to keep param count down.
    r   r€   Tr�   ©r*   r\   é   r*   r†   Fr‡   Ú
skresnet34rŒ   r�   s       r>   r“   r“   Ù   r�   r?   c           	      ó¤   — t          d¬¦  «        }t          t          g d¢t          |¬¦  «        d¬¦  «        }t          d| fi t          |fi |¤Ž¤ŽS )zùConstructs a Select Kernel ResNet-50 model.

    Different from configs in Select Kernel paper or "Compounding the Performance Improvements..." this
    variation splits the input channels to the selective convolutions to keep param count down.
    T©r„   r‘   r†   Fr‡   Ú
skresnet50©r2   r[   ri   r�   s       r>   r–   r–   ç   sl   € õ  Ð&Ñ&Ô&€IÝÝ'°°°ÍÐXaÐIbÑIbÔIbØðñ ô €Jõ ˜L¨*ÐSÐS½¸ZÐ8RÐ8RÈ6Ð8RÐ8RÐSÐSÐSr?   c           
      óª   — t          d¬¦  «        }t          t          g d¢dddt          |¬¦  «        d¬¦  «        }t          d	| fi t          |fi |¤Ž¤ŽS )
zûConstructs a Select Kernel ResNet-50-D model.

    Different from configs in Select Kernel paper or "Compounding the Performance Improvements..." this
    variation splits the input channels to the selective convolutions to keep param count down.
    Tr•   r‘   é    Údeepr†   F)rˆ   r‰   Ú
stem_widthÚ	stem_typeÚavg_downrŠ   rH   Úskresnet50dr—   r�   s       r>   rž   rž   õ   sv   € õ  Ð&Ñ&Ô&€IÝÝ'°°°ÈÐW]ÐhlÝ )Ð,Ñ,Ô,¸UðDñ Dô D€Jõ ˜M¨:ÐTÐT½¸jÐ9SÐ9SÈFÐ9SÐ9SÐTÐTÐTr?   c           	      ó¬   — t          ddd¬¦  «        }t          t          g d¢ddt          |¬¦  «        d¬¦  «        }t          d	| fi t          |fi |¤Ž¤ŽS )
z…Constructs a Select Kernel ResNeXt50-32x4d model. This should be equivalent to
    the SKNet-50 model in the Select Kernel Paper
    g      °?r™   Fr�   r‘   r\   r†   )rˆ   r‰   r   r   rŠ   rH   Úskresnext50_32x4dr—   r�   s       r>   r    r      sy   € õ
 ˜d¨r¸uÐEÑEÔE€IÝÝ'°°°È"ÐYZÝ )Ð,Ñ,Ô,¸UðDñ Dô D€Jõ Ð/°ÐZÐZ½tÀJÐ?YÐ?YÐRXÐ?YÐ?YÐZÐZÐZr?   )F)rj   ) Ú__doc__r_   Útypingr   r   Útorchr   Ú	timm.datar   r   Útimm.layersr	   r
   r   Ú_builderr   Ú	_registryr   r   Úresnetr   rW   r   r[   ri   r{   Údefault_cfgsr‹   r“   r–   rž   r    r/   r?   r>   ú<module>rª      s¼  ðð	ð 	ð €€€Ø !Ð !Ð !Ð !Ð !Ð !Ð !Ð !à Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ AØ AÐ AÐ AÐ AÐ AÐ AÐ AÐ AÐ AÐ AØ *Ð *Ð *Ð *Ð *Ð *Ø <Ð <Ð <Ð <Ð <Ð <Ð <Ð <Ø Ð Ð Ð Ð Ð ðJð Jð Jð Jð J˜2œ9ñ Jô Jð JðZEð Eð Eð Eð E ¤	ñ Eô Eð EðPð ð ð ð	ð 	ð 	ð 	ð %Ð$Ø˜$¨Ð1Ñ1Ô1Ø˜$¨Ð1Ñ1Ô1Ø ˜D™FœFØ!˜TØðñ ô à!% °Ð!8Ñ!8Ô!8ð&ð &ñ ô €ð ð
Tð 
T¨fð 
Tð 
Tð 
Tñ „ð
Tð ð
Tð 
T¨fð 
Tð 
Tð 
Tñ „ð
Tð ð
Tð 
T¨fð 
Tð 
Tð 
Tñ „ð
Tð ð
Uð 
U¨vð 
Uð 
Uð 
Uñ „ð
Uð ð[ð [°Vð [ð [ð [ñ „ð[ð [ð [r?   