
    kh                    !   d Z ddlZddlmZ ddl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mZmZmZm Z m!Z! ddl"m#Z# dd	l$m%Z% dd
l&m'Z' ddl(m)Z)m*Z*m+Z+ g dZ,dde-de-de-de-fdZ. G d dej^                        Z0 G d dej^                        Z1	 	 	 	 dde-de-de-de-de-dee-   dee
ej^                        dej^                  fdZ2	 	 	 	 dde-de-de-de-de-dee-   dee
ej^                        dej^                  fdZ3dde4deee      fdZ5	 	 	 	 	 	 dde	ee
e0   e
e1   f   df   d e	e-df   d!e	e-df   d"e-d#e-d$e-d%e-d&e6d'e4d(e4de	ee	e7ej^                  f      eee7ef      f   fd)Z8 G d* d+ej^                        Z9dd,e7d-e6de9fd.Z:dd/e7dee7ef   fd0Z;dd/e7dee7ef   fd1Z<dd/e7dee7ef   fd2Z=dd/e7dee7ef   fd3Z>dd/e7dee7ef   fd4Z?dd/e7dee7ef   fd5Z@ e*i d6 e=d7d8d9d:d;d<d=>      d? e=d7d@d9d:d;d<d=>      dA e>d7dBC      dD e>d7dEC      dF e?d7dGC      dH e=d7dId=J      dK e>d7dLdLdMd=N      dO e>d7dPC      dQ e>d7dRC      dS e?d7dTd;U      dV e=d7dWC      dX e>d7dLdLdMY      dZ e=d7d[d=J      d\ e=d7d]C      d^ e=d7d_d=J      d` e=d7dad=dbdcdddedfg      dh e>d7diC      i dj e>d7dkd9d:dMd<dfl      dm e>d7dnC      do e?d7dpC      dq e>d7drC      ds e>d7dtC      du e>d7dvC      dw e>d7dxC      dy e>d7dzC      d{ e=d7d|C      d} e<d7d~C      d e=d7dC      d e=d7dC      d e=d7dd=J      d e>d7dLdLd;ddfd=      d e>d7dd=J      d e>d7dd=J      d e?d7dd=J      i d e=d=      d e>d7dC      d e>d7dC      d e>d7dC      d e?d7dC      d e=d7dd=dbdcd;dfde      d e>d7dC      d e>d7dC      d e>d7dC      d e?d7dC      d e=d7dd=dbdcd;dfde      d e=       d e=d7dd=dbdcd;dfde      d e=d7dC      d e;d7ddd      d e;d7ddd      d e;d7ddd      i d e;d7dd9d:d<ddd      d e;d7ddd      d e;d7dd9d:d<ddd      d e;d7ddd      d e;d7dd9d:d<ddd      d e;d7ddd      d e;d7dd9d:d<ddd      d e;d7ddd      d e;d7dd9d:d<ddd      d e=d7dddU      d e>d7dʬC      d e>d7d̬C      d e>d7dάC      d e?d7dЬC      d e=d7dҬC      d e=d7dd=J      d e=       i d e>d7d׬C      d e;d7ddd      d e;d7ddd      d e;d7ddd      d e;d7dd9d:d<ddd      d e;d7dd9d:d<ddd      d e;d7ddd嬮      d e;d7ddd嬮      d e;d7ddd嬮      d e;d7ddd嬮      d e;d7ddd      d e;d7ddd      d e;d7ddd      d e;d7ddd      d e;d7ddd      d e;d7ddd      d e;d7ddd      i d e;d7ddd      d e;d7ddd      d e;d7d dd      d e;d7ddd      d e;d7ddd      d e=d7dd=dbdcd;de      d e<d7d	d;d
      d e<d7dd=d;d      d e<d7dd=d;d      d e<d7dd=dbdcd;de      d e>d7dd=J      d e>d7dd=J      d e?d7dd=J      d e<d7dd=d;d      d e<d7dd=d;d      d e=d=dbd;dc      d e=d7dd=ded d;dfd!      i d" e<d=      d# e<d=      d$ e=       d% e=       d& e>d7d'd;U      d( e>d7d)d;U      d* e?d7d+d;U      d, e=d7d-C      d. e=d=      d/ e=       d0 e=       d1 e=d7d2d=dbdcd;dfde      d3 e=d=dbdc4      d5 e=d=dbdc4      d6 e=d7d7d=J      d8 e=d7d9d=J      d: e=d7d;C      i d< e=       d= e>d7d>C      d? e>d7d@d=J      dA e=d7d=d;dfB      dC e=d7d=d;dfB      dD e=d7d;ddEdedfd=F      dG e=d7d=dfH      dI e;d7dJd=dKdLdfM      dN e;d7dOdJd=d;ded dLdfP	      dQ e=d7dOd=d;dfR      dS e=d7dOd=d;dfR      dT e=d7dOd=d;dfR      dU e=d7dOd=d;dfR      dV e=       dW e=d7dXC      dY e=d=      dZ e=d=      i d[ e=d=      d\ e>d7d]C      d^ e=d=      d_ e>d7d`d=J      da e;d7dbdcddded<dJd=f      dg e;d7dhdid:ddddJd=f      dj e;d7dkdbdcdfdedJd=f      dl e;d7dmdbdcdfdedJd=f      dn e;d7dodbdcdfd!dJd=f      dp e;d7dqddEdfdLdJd=f      dr e;d7dsded dfdtdJd=f      du e@d7dvC      dw e@d7dxC      dy e@d7dzC      d{ e@d7d|C      d} e@d7d~C      d e@d7dd=J      i d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dd=J      d e@d7dC      d e@d7dC      d e@d7dC      d e@d7dC      d e@d7dC      d e@d7dC      d e@d7dd=J      d e;d7dLdLd;dcddd=            ZAe)dd-e6de9fd       ZBe)dd-e6de9fd       ZCe)dd-e6de9fd       ZDe)dd-e6de9fd       ZEe)dd-e6de9fd       ZFe)dd-e6de9fd       ZGe)dd-e6de9fd       ZHe)dd-e6de9fd       ZIe)dd-e6de9fd       ZJe)dd-e6de9fd       ZKe)dd-e6de9fd       ZLe)dd-e6de9fd       ZMe)dd-e6de9fd       ZNe)dd-e6de9fd       ZOe)dd-e6de9fd       ZPe)dd-e6de9fd       ZQe)dd-e6de9fd       ZRe)dd-e6de9fd       ZSe)dd-e6de9fd       ZTe)dd-e6de9fd       ZUe)dd-e6de9fd       ZVe)dd-e6de9fd       ZWe)dd-e6de9fd       ZXe)dd-e6de9fd       ZYe)dd-e6de9fd       ZZe)dd-e6de9fd       Z[e)dd-e6de9fd       Z\e)dd-e6de9fd       Z]e)dd-e6de9fd       Z^e)dd-e6de9fd       Z_e)dd-e6de9fd       Z`e)dd-e6de9fd       Zae)dd-e6de9fd       Zbe)dd-e6de9fd       Zce)dd-e6de9fdÄ       Zde)dd-e6de9fdĄ       Zee)dd-e6de9fdń       Zfe)dd-e6de9fdƄ       Zge)dd-e6de9fdǄ       Zhe)dd-e6de9fdȄ       Zie)dd-e6de9fdɄ       Zje)dd-e6de9fdʄ       Zke)dd-e6de9fd˄       Zle)dd-e6de9fd̄       Zme)dd-e6de9fd̈́       Zne)dd-e6de9fd΄       Zoe)dd-e6de9fdτ       Zpe)dd-e6de9fdЄ       Zqe)dd-e6de9fdф       Zre)dd-e6de9fd҄       Zse)dd-e6de9fdӄ       Zte)dd-e6de9fdԄ       Zue)dd-e6de9fdՄ       Zve)dd-e6de9fdք       Zwe)dd-e6de9fdׄ       Zxe)dd-e6de9fd؄       Zye)dd-e6de9fdل       Zze)dd-e6de9fdڄ       Z{e)dd-e6de9fdۄ       Z|e)dd-e6de9fd܄       Z}e)dd-e6de9fd݄       Z~e)dd-e6de9fdބ       Ze)dd-e6de9fd߄       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Ze)dd-e6de9fd       Z e+ei ddddddddddؓddddddddddddddddddd dddddi dddddddddduddwd	dyd
d{dd}dddddddddddddddddddddddddd	       y(  a*  PyTorch ResNet

This started as a copy of https://github.com/pytorch/vision 'resnet.py' (BSD-3-Clause) with
additional dropout and dynamic global avg/max pool.

ResNeXt, SE-ResNeXt, SENet, and MXNet Gluon stem/downsample variants, tiered stems added by Ross Wightman

Copyright 2019, Ross Wightman
    N)partial)AnyDictListOptionalTupleTypeUnionIMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD)DropBlock2dDropPathAvgPool2dSame
BlurPool2d	LayerTypecreate_attnget_attnget_act_layerget_norm_layercreate_classifier	create_aa	to_ntuple   )build_model_with_cfg)feature_take_indices)checkpoint_seq)register_modelgenerate_default_cfgsregister_model_deprecations)ResNet
BasicBlock
Bottleneckkernel_sizestridedilationreturnc                 (    |dz
  || dz
  z  z   dz  }|S )Nr       )r$   r%   r&   paddings       N/var/www/teggl/fontify/venv/lib/python3.12/site-packages/timm/models/resnet.pyget_paddingr-      s#    
h+/::q@GN    c            !           e Zd ZdZdZdddddddej                  ej                  ddddfdededede	ej                     d	ed
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ej                        de	ej                     ddf  fdZddZdej                  dej                  fdZ xZS )r"   zoBasic residual block for ResNet.

    This is the standard residual block used in ResNet-18 and ResNet-34.
    r   N@   inplanesplanesr%   
downsamplecardinality
base_widthreduce_firstr&   first_dilation	act_layer
norm_layer
attn_layeraa_layer
drop_block	drop_pathr'   c           	      n   t         t        |           |dk(  sJ d       |dk(  sJ d       ||z  }|| j                  z  }|	xs |}	|duxr |dk(  xs |	|k7  }t	        j
                  ||d|rdn||	|	d	      | _         ||      | _        | |       nt	        j                         | _	         |
d
      | _
        t        ||||      | _        t	        j
                  ||d||d      | _         ||      | _        t        ||      | _         |
d
      | _        || _        || _        || _        || _        y)  
        Args:
            inplanes: Input channel dimensionality.
            planes: Used to determine output channel dimensionalities.
            stride: Stride used in convolution layers.
            downsample: Optional downsample layer for residual path.
            cardinality: Number of convolution groups.
            base_width: Base width used to determine output channel dimensionality.
            reduce_first: Reduction factor for first convolution output width of residual blocks.
            dilation: Dilation rate for convolution layers.
            first_dilation: Dilation rate for first convolution layer.
            act_layer: Activation layer class.
            norm_layer: Normalization layer class.
            attn_layer: Attention layer class.
            aa_layer: Anti-aliasing layer class.
            drop_block: DropBlock layer class.
            drop_path: Optional DropPath layer instance.
        r   z)BasicBlock only supports cardinality of 1r0   z/BasicBlock does not support changing base widthNr)      F)r$   r%   r+   r&   biasTinplacechannelsr%   enable)r$   r+   r&   rA   )superr"   __init__	expansionnnConv2dconv1bn1Identityr<   act1r   aaconv2bn2r   seact2r3   r%   r&   r=   )selfr1   r2   r%   r3   r4   r5   r6   r&   r7   r8   r9   r:   r;   r<   r=   first_planes	outplanesuse_aa	__class__s                      r,   rH   zBasicBlock.__init__)   sC   H 	j$(*aL!LLRR!RR-T^^+	'38%U6Q;+T.H:TYYlv!6[i#%1
 l+*4*@*,bkkmd+	H|FSYZYY)Hx^ce
i(j)4d+	$ "r.   c                     t        | j                  dd      4t        j                  j	                  | j                  j
                         yyzLInitialize the last batch norm layer weights to zero for better convergence.weightN)getattrrR   rJ   initzeros_r\   rU   s    r,   zero_init_lastzBasicBlock.zero_init_lastj   2    488Xt,8GGNN488??+ 9r.   xc                    |}| j                  |      }| j                  |      }| j                  |      }| j                  |      }| j	                  |      }| j                  |      }| j                  |      }| j                  | j                  |      }| j                  | j                  |      }| j                  | j                  |      }||z  }| j                  |      }|S N)rL   rM   r<   rO   rP   rQ   rR   rS   r=   r3   rT   rU   rc   shortcuts      r,   forwardzBasicBlock.forwardo   s    JJqMHHQKOOAIIaLGGAJJJqMHHQK77
A>>%q!A??&x0H	XIIaLr.   r'   N__name__
__module____qualname____doc__rI   rJ   ReLUBatchNorm2dintr   Moduler	   rH   ra   torchTensorrh   __classcell__rY   s   @r,   r"   r"   "   sQ    I .2   !,0)+*,..482648-1!?#?# ?# 	?#
 !+?# ?# ?# ?# ?# %SM?# BII?# RYY?# !bii1?# tBII/?# !bii1?#   		*!?#" 
#?#B,
 %,, r.   r"   c            !           e Zd ZdZdZdddddddej                  ej                  ddddfdededed	e	ej                     d
ed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ej                        de	ej                     ddf  fdZddZdej                  dej                  fdZ xZS )r#   z{Bottleneck residual block for ResNet.

    This is the bottleneck block used in ResNet-50, ResNet-101, and ResNet-152.
       r   Nr0   r1   r2   r%   r3   r4   r5   r6   r&   r7   r8   r9   r:   r;   r<   r=   r'   c           
         t         t        |           t        t	        j
                  ||dz  z        |z        }||z  }|| j                  z  }|	xs |}	|duxr |dk(  xs |	|k7  }t        j                  ||dd      | _	         ||      | _
         |
d      | _        t        j                  ||d	|rdn||	|	|d
      | _         ||      | _        | |       nt        j                         | _         |
d      | _        t#        ||||      | _        t        j                  ||dd      | _         ||      | _        t+        ||      | _         |
d      | _        || _        || _        || _        || _        y)r?   r0   Nr)   r   F)r$   rA   TrB   r@   )r$   r%   r+   r&   groupsrA   rD   )rG   r#   rH   rq   mathfloorrI   rJ   rK   rL   rM   rO   rQ   rR   rN   r<   rT   r   rP   conv3bn3r   rS   act3r3   r%   r&   r=   )rU   r1   r2   r%   r3   r4   r5   r6   r&   r7   r8   r9   r:   r;   r<   r=   widthrV   rW   rX   rY   s                       r,   rH   zBottleneck.__init__   sc   H 	j$(*DJJvb9:[HI,T^^+	'38%U6Q;+T.H:TYYx15Q
l+d+	YY%QFq"^KV[]
 e$*4*@*,bkkmd+	HuVFSYYuiQUK
i(j)4d+	$ "r.   c                     t        | j                  dd      4t        j                  j	                  | j                  j
                         yyr[   )r]   r~   rJ   r^   r_   r\   r`   s    r,   ra   zBottleneck.zero_init_last   rb   r.   rc   c                 8   |}| j                  |      }| j                  |      }| j                  |      }| j                  |      }| j	                  |      }| j                  |      }| j                  |      }| j                  |      }| j                  |      }| j                  |      }| j                  | j                  |      }| j                  | j                  |      }| j                  | j                  |      }||z  }| j                  |      }|S re   )rL   rM   rO   rQ   rR   r<   rT   rP   r}   r~   rS   r=   r3   r   rf   s      r,   rh   zBottleneck.forward   s    JJqMHHQKIIaLJJqMHHQKOOAIIaLGGAJJJqMHHQK77
A>>%q!A??&x0H	XIIaLr.   ri   rj   rv   s   @r,   r#   r#      sb    I .2   !,0)+*,..482648-1!A#A# A# 	A#
 !+A# A# A# A# A# %SMA# BIIA# RYYA# !bii1A# tBII/A# !bii1A#   		*!A#" 
#A#F,
 %,, r.   r#   in_channelsout_channelsr7   r9   c                     |xs t         j                  }|dk(  r|dk(  rdn|}|dkD  r|xs |nd}t        |||      }t        j                  t        j                  | |||||d       ||      g S )Nr   F)r%   r+   r&   rA   )rJ   rp   r-   
SequentialrK   )r   r   r$   r%   r&   r7   r9   ps           r,   downsample_convr      s     -r~~J{x1}!+K5@1_n0!NK8A==
		{61Welq	s<   r.   c                 :   |xs t         j                  }|dk(  r|nd}|dk(  r|dk(  rt        j                         }n,|dk(  r|dkD  rt        nt         j                  }	 |	d|dd      }t        j
                  |t        j                  | |dddd       ||      g S )Nr   r)   TF)	ceil_modecount_include_padr   r%   r+   rA   )rJ   rp   rN   r   	AvgPool2dr   rK   )
r   r   r$   r%   r&   r7   r9   
avg_stridepoolavg_pool_fns
             r,   downsample_avgr     s     -r~~J#q=aJ{x1}{{}'1Q8a<mR\\1jDER==
		+|Qq!%P<   r.   	drop_probc           	      d    dd| rt        t        | dd      nd| rt        t        | dd      gS dgS )zCreate DropBlock layer instances for each stage.

    Args:
        drop_prob: Drop probability for DropBlock.

    Returns:
        List of DropBlock partial instances or None for each stage.
    N         ?)r   
block_sizegamma_scaler@         ?)r   r   )r   s    r,   drop_blocksr   #  sK     	dU^yQDQdhU^yQDQj j eij jr.   	block_fns.rE   block_repeatsr1   r6   output_stridedown_kernel_sizeavg_downdrop_block_ratedrop_path_ratec
                    g }g }t        |      }d}d}dx}}t        t        | ||t        |                  D ]O  \  }\  }}}}d|dz    }|dk(  rdnd}||k\  r||z  }d}n||z  }d}|dk7  s|||j                  z  k7  rFt        |||j                  z  |||||
j                  d            }|rt        di |n
t        di |}t        d|||d	|
}g }t        |      D ]c  }|dk(  r|nd}|dk(  r|nd}|	|z  |dz
  z  }|j                   |||||f||d
kD  rt        |      ndd|       |}||j                  z  }|dz  }e |j                  |t        j                  | f       |j                  t        |||             R ||fS )a  Create ResNet stages with specified block configurations.

    Args:
        block_fns: Block class to use for each stage.
        channels: Number of channels for each stage.
        block_repeats: Number of blocks to repeat for each stage.
        inplanes: Number of input channels.
        reduce_first: Reduction factor for first convolution in each stage.
        output_stride: Target output stride of network.
        down_kernel_size: Kernel size for downsample layers.
        avg_down: Use average pooling for downsample.
        drop_block_rate: DropBlock drop rate.
        drop_path_rate: Drop path rate for stochastic depth.
        **kwargs: Additional arguments passed to block constructors.

    Returns:
        Tuple of stage modules list and feature info list.
    r   rx   r   layerr)   Nr9   )r   r   r$   r%   r&   r7   r9   )r6   r&   r<           )r7   r=   num_chs	reductionmoduler*   )sum	enumeratezipr   rI   dictgetr   r   rangeappendr   rJ   r   )r   rE   r   r1   r6   r   r   r   r   r   kwargsstagesfeature_infonet_num_blocksnet_block_idx
net_strider&   prev_dilation	stage_idxblock_fnr2   
num_blocksdb
stage_namer%   r3   down_kwargsblock_kwargsblocks	block_idx	block_dprs                                  r,   make_blocksr   2  s%   > FL'NMJ  H}9B3yRZ\ikv  xG  lH  DI  :J *]5	5Hfj"Y]O,
1n!&HF& J
Q;(fx/A/A&AA$#h&8&88,!,!::l3K ;C6+6HfZeHfJbUWb[abz* 	I'0A~4J(A~V1F&6.1:LMIMM(	
  -1:R(9-T   %M 2 22HQM	" 	z2==&#9:;DZPZ[\U*]X <r.   c            2           e Zd ZdZddddddddd	ddd	d
ej
                  ej                  ddddddfdeee	f   de
edf   dededededededededededededee
edf      deded eeej"                        d!ed"ed#ed$ed%eeeef      f. fd&Zej.                  j0                  d>d$ed'dfd(       Zej.                  j0                  d?d)ed'eeef   fd*       Zej.                  j0                  d>d+ed'dfd,       Zej.                  j0                  d?d-ed'eeej"                  f   fd.       Zd@deded'dfd/Z	 	 	 	 	 dAd0ej<                  d1eeeee   f      d2ed3ed4ed5ed'eeej<                     e
ej<                  eej<                     f   f   fd6Z 	 	 	 dBd1eeee   f   d7ed8ed'ee   fd9Z!d0ej<                  d'ej<                  fd:Z"d?d0ej<                  d;ed'ej<                  fd<Z#d0ej<                  d'ej<                  fd=Z$ xZ%S )Cr!   a  ResNet / ResNeXt / SE-ResNeXt / SE-Net

    This class implements all variants of ResNet, ResNeXt, SE-ResNeXt, and SENet that
      * have > 1 stride in the 3x3 conv layer of bottleneck
      * have conv-bn-act ordering

    This ResNet impl supports a number of stem and downsample options based on the v1c, v1d, v1e, and v1s
    variants included in the MXNet Gluon ResNetV1b model. The C and D variants are also discussed in the
    'Bag of Tricks' paper: https://arxiv.org/pdf/1812.01187. The B variant is equivalent to torchvision default.

    ResNet variants (the same modifications can be used in SE/ResNeXt models as well):
      * normal, b - 7x7 stem, stem_width = 64, same as torchvision ResNet, NVIDIA ResNet 'v1.5', Gluon v1b
      * c - 3 layer deep 3x3 stem, stem_width = 32 (32, 32, 64)
      * d - 3 layer deep 3x3 stem, stem_width = 32 (32, 32, 64), average pool in downsample
      * e - 3 layer deep 3x3 stem, stem_width = 64 (64, 64, 128), average pool in downsample
      * s - 3 layer deep 3x3 stem, stem_width = 64 (64, 64, 128)
      * t - 3 layer deep 3x3 stem, stem width = 32 (24, 48, 64), average pool in downsample
      * tn - 3 layer deep 3x3 stem, stem width = 32 (24, 32, 64), average pool in downsample

    ResNeXt
      * normal - 7x7 stem, stem_width = 64, standard cardinality and base widths
      * same c,d, e, s variants as ResNet can be enabled

    SE-ResNeXt
      * normal - 7x7 stem, stem_width = 64
      * same c, d, e, s variants as ResNet can be enabled

    SENet-154 - 3 layer deep 3x3 stem (same as v1c-v1s), stem_width = 64, cardinality=64,
        reduction by 2 on width of first bottleneck convolution, 3x3 downsample convs after first block
      r@       avgr   r0    F)r0         i   Nr   Tblocklayers.num_classesin_chansr   global_poolr4   r5   
stem_width	stem_typereplace_stem_poolblock_reduce_firstr   r   rE   r8   r9   r;   	drop_rater   r   ra   
block_argsc                    t         t        |           |xs
 t               }|dv sJ || _        || _        d| _        t        |      }t        |      }d|
v }|r|	dz  nd}|r|	|	f}d|
v r
d|	dz  z  |	f}t        j                  t        j                  ||d	   ddd
d       ||d	          |d      t        j                  |d	   |d
   dd
d
d       ||d
          |d      t        j                  |d
   |dd
d
d      g | _        n t        j                  ||dddd      | _         ||      | _         |d      | _        t        |dd      g| _        |r`t        j                  t!        dt        j                  ||d|rd
ndd
d      |t#        ||d      nd ||       |d      g       | _        n|`t'        |t        j(                        r |d      | _        nUt        j                  t        j*                  dd
d
       ||d      g | _        nt        j*                  ddd
      | _         t-        t/        |            |      }t1        ||||f|||||||||||d|\  }}|D ]  } | j2                  |   | j                  j5                  |       |d   |d   j6                  z  x| _        | _        t=        | j8                  | j                  |      \  | _        | _         | jC                  |       y)a	  
        Args:
            block (nn.Module): class for the residual block. Options are BasicBlock, Bottleneck.
            layers (List[int]) : number of layers in each block
            num_classes (int): number of classification classes (default 1000)
            in_chans (int): number of input (color) channels. (default 3)
            output_stride (int): output stride of the network, 32, 16, or 8. (default 32)
            global_pool (str): Global pooling type. One of 'avg', 'max', 'avgmax', 'catavgmax' (default 'avg')
            cardinality (int): number of convolution groups for 3x3 conv in Bottleneck. (default 1)
            base_width (int): bottleneck channels factor. `planes * base_width / 64 * cardinality` (default 64)
            stem_width (int): number of channels in stem convolutions (default 64)
            stem_type (str): The type of stem (default ''):
                * '', default - a single 7x7 conv with a width of stem_width
                * 'deep' - three 3x3 convolution layers of widths stem_width, stem_width, stem_width * 2
                * 'deep_tiered' - three 3x3 conv layers of widths stem_width//4 * 3, stem_width, stem_width * 2
            replace_stem_pool (bool): replace stem max-pooling layer with a 3x3 stride-2 convolution
            block_reduce_first (int): Reduction factor for first convolution output width of residual blocks,
                1 for all archs except senets, where 2 (default 1)
            down_kernel_size (int): kernel size of residual block downsample path,
                1x1 for most, 3x3 for senets (default: 1)
            avg_down (bool): use avg pooling for projection skip connection between stages/downsample (default False)
            act_layer (str, nn.Module): activation layer
            norm_layer (str, nn.Module): normalization layer
            aa_layer (nn.Module): anti-aliasing layer
            drop_rate (float): Dropout probability before classifier, for training (default 0.)
            drop_path_rate (float): Stochastic depth drop-path rate (default 0.)
            drop_block_rate (float): Drop block rate (default 0.)
            zero_init_last (bool): zero-init the last weight in residual path (usually last BN affine weight)
            block_args (dict): Extra kwargs to pass through to block module
        )      r   Fdeepr)   r0   tieredr@   rx   r   r   r   TrB      )r$   r%   r+   rA   rO   r   N)rE   r%   )r$   r%   r+   )r4   r5   r   r6   r   r   r8   r9   r;   r   r   	pool_type)ra   )"rG   r!   rH   r   r   r   grad_checkpointingr   r   rJ   r   rK   rL   rM   rO   r   filterr   maxpool
issubclassr   	MaxPool2dr   lenr   
add_moduleextendrI   num_featureshead_hidden_sizer   r   fcinit_weights) rU   r   r   r   r   r   r   r4   r5   r   r   r   r   r   r   rE   r8   r9   r;   r   r   r   ra   r   	deep_stemr1   stem_chsr   stage_modulesstage_feature_infostagerY   s                                   r,   rH   zResNet.__init__  s3   p 	fd$&)46
+++&""'!),	#J/
 i'	%.:>B"J/H9$q1:>		(HQK1aeT8A;'$'		(1+x{AaQVW8A;'$'		(1+x1aeT)V WDJ 8X1QXY`efDJh'd+	!(aOP ==&		(HaX1VW^cdDLDX	(Xa@^b8$$'	8 + DL #h5#+A;DL#%==1aH (1=3? $@DL  "||!QO -Ic(m,U3	,7	-

 $!'+-!+)-
  !-
))$ # 	$EDOOU#	$  !34 5=RL9R=CZCZ4ZZD1$5d6G6GIYIYep$q!$'8r.   r'   c                 ,   | j                         D ]L  \  }}t        |t        j                        s!t        j                  j                  |j                  dd       N |r3| j                         D ]  }t        |d      s|j                          ! yy)zInitialize model weights.

        Args:
            zero_init_last: Zero-initialize the last BN in each residual branch.
        fan_outrelu)modenonlinearityra   N)
named_modules
isinstancerJ   rK   r^   kaiming_normal_r\   moduleshasattrra   )rU   ra   nms       r,   r   zResNet.init_weights.  s     &&( 	WDAq!RYY'''yv'V	W \\^ '1./$$&' r.   coarsec                 (    t        d|rdnd      }|S )zCreate regex patterns for parameter grouping.

        Args:
            coarse: Use coarse (stage-level) or fine (block-level) grouping.

        Returns:
            Dictionary mapping group names to regex patterns.
        z^conv1|bn1|maxpoolz^layer(\d+)z^layer(\d+)\.(\d+))stemr   )r   )rU   r   matchers      r,   group_matcherzResNet.group_matcher=  s     1F.Xmnr.   rF   c                     || _         y)zEnable or disable gradient checkpointing.

        Args:
            enable: Whether to enable gradient checkpointing.
        N)r   )rU   rF   s     r,   set_grad_checkpointingzResNet.set_grad_checkpointingJ  s     #)r.   	name_onlyc                 "    |rdS | j                   S )zGet the classifier module.

        Args:
            name_only: Return classifier module name instead of module.

        Returns:
            Classifier module or name.
        r   )r   )rU   r   s     r,   get_classifierzResNet.get_classifierS  s     !t-dgg-r.   c                 p    || _         t        | j                  | j                   |      \  | _        | _        y)zReset the classifier head.

        Args:
            num_classes: Number of classes for new classifier.
            global_pool: Global pooling type.
        r   N)r   r   r   r   r   )rU   r   r   s      r,   reset_classifierzResNet.reset_classifier_  s1     '$5d6G6GIYIYep$q!$'r.   rc   indicesnorm
stop_early
output_fmtintermediates_onlyc                 z   |dv sJ d       g }t        d|      \  }}	d}
| j                  |      }| j                  |      }| j                  |      }|
|v r|j	                  |       | j                  |      }d}|r|d|	 }|D ]/  }|
dz  }
 t        | |      |      }|
|v s|j	                  |       1 |r|S ||fS )aJ  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 compatible 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:
            Features and list of intermediate features or just intermediate features.
        )NCHWzOutput shape must be NCHW.r   r   layer1layer2layer3layer4Nr   )r   rL   rM   rO   r   r   r]   )rU   rc   r  r  r  r  r	  intermediatestake_indices	max_indexfeat_idxlayer_namesr   s                r,   forward_intermediateszResNet.forward_intermediatesi  s    , Y&D(DD&"6q'"Bi JJqMHHQKIIaL|#  #LLO>%jy1K 	(AMH a #A<'$$Q'		(   -r.   
prune_norm
prune_headc                     t        d|      \  }}d}||d }|D ]!  }t        | |t        j                                # |r| j	                  dd       |S )aF  Prune layers not required for specified intermediates.

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

        Returns:
            List of indices that were kept.
        r   r  Nr   r   )r   setattrrJ   rN   r  )rU   r  r  r  r  r  r  r   s           r,   prune_intermediate_layersz ResNet.prune_intermediate_layers  s`      #7q'"Bi>!)*- 	,AD!R[[]+	,!!!R(r.   c                    | j                  |      }| j                  |      }| j                  |      }| j                  |      }| j                  rZt
        j                  j                         s<t        | j                  | j                  | j                  | j                  g|d      }|S | j                  |      }| j                  |      }| j                  |      }| j                  |      }|S )z/Forward pass through feature extraction layers.T)flatten)rL   rM   rO   r   r   rs   jitis_scriptingr   r  r  r  r  rU   rc   s     r,   forward_featureszResNet.forward_features  s    JJqMHHQKIIaLLLO""599+A+A+CT[[$++t{{SUV`deA 	 AAAAAAAAr.   
pre_logitsc                     | j                  |      }| j                  r5t        j                  |t	        | j                        | j
                        }|r|S | j                  |      S )zForward pass through classifier head.

        Args:
            x: Feature tensor.
            pre_logits: Return features before final classifier layer.

        Returns:
            Output tensor.
        )r   training)r   r   Fdropoutfloatr$  r   )rU   rc   r"  s      r,   forward_headzResNet.forward_head  sO     Q>>		!uT^^4t}}MAq.DGGAJ.r.   c                 J    | j                  |      }| j                  |      }|S )zForward pass.)r!  r(  r   s     r,   rh   zResNet.forward  s'    !!!$a r.   )TF)r   )NFFr  F)r   FT)&rk   rl   rm   rn   rJ   ro   rp   r
   r"   r#   r   rq   strboolr   r   r	   rr   r'  r   r   rH   rs   r  ignorer   r   r   r  r  rt   r   r  r  r!  r(  rh   ru   rv   s   @r,   r!   r!     s   F  $!#$   &+&'$%"2E#%77$&NN26"$&%'#'371F9Z/0F9 #s(OF9 	F9
 F9 F9 F9 F9 F9 F9 F9  $F9 !$F9 "F9 F9  uS#X/!F9" !#F9$ "%F9& tBII/'F9( )F9* "+F9, #-F9. !/F90 !c3h01F9P YY'4 '4 ' ' YY
D 
T#s(^ 
 
 YY)T )T ) ) YY	. 	.sBII~9N 	. 	.rC rc rd r 8<$$',/ ||/  eCcN34/  	/ 
 /  /  !%/  
tELL!5tELL7I)I#JJ	K/ f ./$#	3S	>*  	
 
c2%,, 5<<  /ell / / / %,, r.   r!   variant
pretrainedc                 &    t        t        | |fi |S )zCreate a ResNet model.

    Args:
        variant: Model variant name.
        pretrained: Load pretrained weights.
        **kwargs: Additional model arguments.

    Returns:
        ResNet model instance.
    )r   r!   )r.  r/  r   s      r,   _create_resnetr1    s      FvFFr.   urlc                 0    | dddddt         t        ddd
|S )	z1Create a default configuration for ResNet models.r   r@      r5  )r   r   g      ?bilinearrL   r   )
r2  r   
input_size	pool_sizecrop_pctinterpolationmeanstd
first_conv
classifierr   r2  r   s     r,   _cfgr@    s2     =vJ%.BT  r.   c           	      4    t        dd| it        ddifi |S )z2Create a configuration with bicubic interpolation.r2  r:  bicubicr*   r@  r   r?  s     r,   _tcfgrD    s$    HCH4) <GGHHr.   c                 :    t        dd| it        dddddfi |S )z4Create a configuration for models trained with timm.r2  rB  r@      rG  ffffff?3https://github.com/huggingface/pytorch-image-models)r:  test_input_sizetest_crop_pct
origin_urlr*   rC  r?  s     r,   _ttcfgrM    s<     C 4"}W[K!  
  r.   c                 >    t        d	d| it        dddddddfi |S )
z,Create a configuration for ResNet-RS models.r2  rB  rH  rF  r   rI  arXiv:2110.00476)r:  r9  rJ  rK  rL  	paper_idsr*   rC  r?  s     r,   _rcfgrQ     sA     C 4"ilKZl!  
  r.   c                 B    t        d
d| it        ddddddddd	fi |S )z?Create a configuration for ResNet-RS models with 160x160 input.r2  rB  r@      rT  r   r   rH  r4  rI  rO  )r:  r7  r8  r9  rJ  rK  rL  rP  r*   rC  r?  s     r,   _r3cfgrV    sE     C 4"-f]TKZl!  
	  r.   c           	      6    t        dd| it        dddfi |S )z3Create a configuration for Gluon pretrained models.r2  rB  z1https://cv.gluon.ai/model_zoo/classification.html)r:  rL  r*   rC  r?  s     r,   _gcfgrX    s7     C 4"I!  
  r.   zresnet10t.c3_in1kztimm/zrhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet10t_176_c3-f3215ab1.pth)r@      rY  )   rZ  rH  r4  zconv1.0)	hf_hub_idr2  r7  r8  rK  rJ  r=  zresnet14t.c3_in1kzrhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet14t_176_c3-c4ed2c37.pthzresnet18.a1_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet18_a1_0-d63eafa0.pth)r[  r2  zresnet18.a2_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet18_a2_0-b61bd467.pthzresnet18.a3_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet18_a3_0-40c531c8.pthzresnet18d.ra2_in1kzkhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet18d_ra2-48a79e06.pth)r[  r2  r=  zresnet18d.ra4_e3600_r224_in1k)      ?r\  r\  g?)r[  r;  r<  r9  r=  zresnet34.a1_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet34_a1_0-46f8f793.pthzresnet34.a2_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet34_a2_0-82d47d71.pthzresnet34.a3_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet34_a3_0-a20cabb6.pth)r[  r2  r9  zresnet34.bt_in1kzfhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet34-43635321.pthzresnet34.ra4_e3600_r224_in1k)r[  r;  r<  r9  zresnet34d.ra2_in1kzkhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet34d_ra2-f8dcfcaf.pthzresnet26.bt_in1kzfhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet26-9aa10e23.pthzresnet26d.bt_in1kzghttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet26d-69e92c46.pthzresnet26t.ra2_in1kzthttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-attn-weights/resnet26t_256_ra2-6f6fa748.pth)r@   r   r   )r   r   gGz?)r@   @  r]  r   )r[  r2  r=  r7  r8  r9  rJ  rK  zresnet50.a1_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_a1_0-14fe96d1.pthzresnet50.a1h_in1kzuhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_a1h2_176-001a1197.pth)r[  r2  r7  r8  r9  rJ  rK  zresnet50.a2_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_a2_0-a2746f79.pthzresnet50.a3_in1kzqhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_a3_0-59cae1ef.pthzresnet50.b1k_in1kzphttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_b1k-532a802a.pthzresnet50.b2k_in1kzphttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_b2k-1ba180c1.pthzresnet50.c1_in1kzohttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_c1-5ba5e060.pthzresnet50.c2_in1kzohttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_c2-d01e05b2.pthzresnet50.d_in1kznhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_d-f39db8af.pthzresnet50.ram_in1kzlhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/resnet50_ram-a26f946b.pthzresnet50.am_in1kzkhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/resnet50_am-6c502b37.pthzresnet50.ra_in1kzkhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/resnet50_ra-85ebb6e5.pthzresnet50.bt_in1kzkhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/rw_resnet50-86acaeed.pthzresnet50d.ra2_in1kzkhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet50d_ra2-464e36ba.pthzresnet50d.ra4_e3600_r224_in1krF  )r[  r;  r<  r9  rJ  rK  r=  zresnet50d.a1_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50d_a1_0-e20cff14.pthzresnet50d.a2_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50d_a2_0-a3adc64d.pthzresnet50d.a3_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50d_a3_0-403fdfad.pthzresnet50t.untrained)r=  zresnet101.a1h_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet101_a1h-36d3f2aa.pthzresnet101.a1_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet101_a1_0-cdcb52a9.pthzresnet101.a2_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet101_a2_0-6edb36c7.pthzresnet101.a3_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet101_a3_0-1db14157.pthzresnet101d.ra2_in1kzlhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet101d_ra2-2803ffab.pth)r[  r2  r=  r7  r8  r9  rK  rJ  zresnet152.a1h_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet152_a1h-dc400468.pthzresnet152.a1_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet152_a1_0-2eee8a7a.pthzresnet152.a2_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet152_a2_0-b4c6978f.pthzresnet152.a3_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet152_a3_0-134d4688.pthzresnet152d.ra2_in1kzlhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet152d_ra2-5cac0439.pthzresnet200.untrainedzresnet200d.ra2_in1kzlhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnet200d_ra2-bdba9bf9.pthzwide_resnet50_2.racm_in1kzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/wide_resnet50_racm-8234f177.pthzresnet18.tv_in1kz9https://download.pytorch.org/models/resnet18-f37072fd.pthzbsd-3-clausez!https://github.com/pytorch/vision)r[  r2  licenserL  zresnet34.tv_in1kz9https://download.pytorch.org/models/resnet34-b627a593.pthzresnet50.tv_in1kz9https://download.pytorch.org/models/resnet50-0676ba61.pthzresnet50.tv2_in1kz9https://download.pytorch.org/models/resnet50-11ad3fa6.pthgzG?)r[  r2  r7  r8  rJ  rK  r^  rL  zresnet101.tv_in1kz:https://download.pytorch.org/models/resnet101-63fe2227.pthzresnet101.tv2_in1kz:https://download.pytorch.org/models/resnet101-cd907fc2.pthzresnet152.tv_in1kz:https://download.pytorch.org/models/resnet152-394f9c45.pthzresnet152.tv2_in1kz:https://download.pytorch.org/models/resnet152-f82ba261.pthzwide_resnet50_2.tv_in1kz@https://download.pytorch.org/models/wide_resnet50_2-95faca4d.pthzwide_resnet50_2.tv2_in1kz@https://download.pytorch.org/models/wide_resnet50_2-9ba9bcbe.pthzwide_resnet101_2.tv_in1kzAhttps://download.pytorch.org/models/wide_resnet101_2-32ee1156.pthzwide_resnet101_2.tv2_in1kzAhttps://download.pytorch.org/models/wide_resnet101_2-d733dc28.pthzresnet50_gn.a1h_in1kzrhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnet50_gn_a1h2-8fe6c4d0.pthzresnext50_32x4d.a1h_in1kzuhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnext50_32x4d_a1h-0146ab0a.pthzresnext50_32x4d.a1_in1kzxhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnext50_32x4d_a1_0-b5a91a1d.pthzresnext50_32x4d.a2_in1kzxhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnext50_32x4d_a2_0-efc76add.pthzresnext50_32x4d.a3_in1kzxhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/resnext50_32x4d_a3_0-3e450271.pthzresnext50_32x4d.ra_in1kzrhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/resnext50_32x4d_ra-d733960d.pthzresnext50d_32x4d.bt_in1kznhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnext50d_32x4d-103e99f8.pthzresnext101_32x4d.untrainedzresnext101_64x4d.c1_in1kzthttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tpu-weights/resnext101_64x4d_c-0d0e0cc0.pthzresnext50_32x4d.tv_in1kz@https://download.pytorch.org/models/resnext50_32x4d-7cdf4587.pthzresnext101_32x8d.tv_in1kzAhttps://download.pytorch.org/models/resnext101_32x8d-8ba56ff5.pthzresnext101_64x4d.tv_in1kzAhttps://download.pytorch.org/models/resnext101_64x4d-173b62eb.pthzresnext50_32x4d.tv2_in1kz@https://download.pytorch.org/models/resnext50_32x4d-1a0047aa.pthzresnext101_32x8d.tv2_in1kzAhttps://download.pytorch.org/models/resnext101_32x8d-110c445d.pthz$resnext101_32x8d.fb_wsl_ig1b_ft_in1kzChttps://download.pytorch.org/models/ig_resnext101_32x8-c38310e5.pthzcc-by-nc-4.0z.https://github.com/facebookresearch/WSL-Imagesz%resnext101_32x16d.fb_wsl_ig1b_ft_in1kzDhttps://download.pytorch.org/models/ig_resnext101_32x16-c6f796b0.pthz%resnext101_32x32d.fb_wsl_ig1b_ft_in1kzDhttps://download.pytorch.org/models/ig_resnext101_32x32-e4b90b00.pthz%resnext101_32x48d.fb_wsl_ig1b_ft_in1kzDhttps://download.pytorch.org/models/ig_resnext101_32x48-3e41cc8a.pthz resnet18.fb_ssl_yfcc100m_ft_in1kzdhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnet18-d92f0530.pthzEhttps://github.com/facebookresearch/semi-supervised-ImageNet1K-modelsz resnet50.fb_ssl_yfcc100m_ft_in1kzdhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnet50-08389792.pthz'resnext50_32x4d.fb_ssl_yfcc100m_ft_in1kzjhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnext50_32x4-ddb3e555.pthz(resnext101_32x4d.fb_ssl_yfcc100m_ft_in1kzkhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnext101_32x4-dc43570a.pthz(resnext101_32x8d.fb_ssl_yfcc100m_ft_in1kzkhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnext101_32x8-2cfe2f8b.pthz)resnext101_32x16d.fb_ssl_yfcc100m_ft_in1kzlhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_supervised_resnext101_32x16-15fffa57.pthzresnet18.fb_swsl_ig1b_ft_in1kzkhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnet18-118f1556.pthzresnet50.fb_swsl_ig1b_ft_in1kzkhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnet50-16a12f1b.pthz$resnext50_32x4d.fb_swsl_ig1b_ft_in1kzqhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnext50_32x4-72679e44.pthz%resnext101_32x4d.fb_swsl_ig1b_ft_in1kzrhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnext101_32x4-3f87e46b.pthz%resnext101_32x8d.fb_swsl_ig1b_ft_in1kzrhttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnext101_32x8-b4712904.pthz&resnext101_32x16d.fb_swsl_ig1b_ft_in1kzshttps://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnext101_32x16-f3559a9c.pthzecaresnet26t.ra2_in1kznhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecaresnet26t_ra2-46609757.pth)r[  r2  r=  r7  r8  rK  rJ  zecaresnetlight.miil_in1kzlhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/ecaresnetlight-75a9c627.pth)r[  r2  rK  rJ  zecaresnet50d.miil_in1kzjhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/ecaresnet50d-93c81e3b.pth)r[  r2  r=  rK  rJ  zecaresnet50d_pruned.miil_in1kzlhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/ecaresnet50d_p-e4fa23c2.pthzecaresnet50t.ra2_in1kznhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecaresnet50t_ra2-f7ac63c4.pthzecaresnet50t.a1_in1kzuhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/ecaresnet50t_a1_0-99bd76a8.pthzecaresnet50t.a2_in1kzuhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/ecaresnet50t_a2_0-b1c7b745.pthzecaresnet50t.a3_in1kzuhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/ecaresnet50t_a3_0-8cc311f1.pthzecaresnet101d.miil_in1kzkhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/ecaresnet101d-153dad65.pthzecaresnet101d_pruned.miil_in1kzmhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/ecaresnet101d_p-9e74cb91.pthzecaresnet200d.untrained)r=  r7  r9  r8  zecaresnet269d.ra2_in1kzshttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ecaresnet269d_320_ra2-7baa55cb.pth)
   r_  )r@   `  r`  zecaresnext26t_32x4d.untrainedzecaresnext50t_32x4d.untrainedzseresnet18.untrainedzseresnet34.untrainedzseresnet50.a1_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/seresnet50_a1_0-ffa00869.pthzseresnet50.a2_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/seresnet50_a2_0-850de0d9.pthzseresnet50.a3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-rsb-weights/seresnet50_a3_0-317ecd56.pthzseresnet50.ra2_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet50_ra_224-8efdb4bb.pthzseresnet50t.untrainedzseresnet101.untrainedzseresnet152.untrainedzseresnet152d.ra2_in1kznhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet152d_ra2-04464dd2.pthzseresnet200d.untrained)r=  r7  r8  zseresnet269d.untrainedzseresnext26d_32x4d.bt_in1kzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnext26d_32x4d-80fa48a3.pthzseresnext26t_32x4d.bt_in1kzqhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnext26tn_32x4d-569cb627.pthzseresnext50_32x4d.racm_in1kzthttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnext50_32x4d_racm-a304a460.pthzseresnext101_32x4d.untrainedzseresnext101_32x8d.ah_in1kzwhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tpu-weights/seresnext101_32x8d_ah-e6bc4c0a.pthzseresnext101d_32x8d.ah_in1kzxhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tpu-weights/seresnext101d_32x8d_ah-191d7b94.pthzresnetaa50d.sw_in12k_ft_in1k)r[  r=  r9  rK  zresnetaa101d.sw_in12k_ft_in1kz*seresnextaa101d_32x8d.sw_in12k_ft_in1k_288)	   ra  )r[  r9  r7  r8  rJ  rK  r=  z&seresnextaa101d_32x8d.sw_in12k_ft_in1k)r[  r=  rK  z*seresnextaa201d_32x8d.sw_in12k_ft_in1k_384rB  )   rb  )r@     rc  )r[  r:  r=  r8  r7  r9  zseresnextaa201d_32x8d.sw_in12ki-.  )	r[  r   r:  r=  r9  r7  r8  rJ  rK  zresnetaa50d.sw_in12k)r[  r   r=  r9  rK  zresnetaa50d.d_in12kzresnetaa101d.sw_in12kzseresnextaa101d_32x8d.sw_in12kzresnetblur18.untrainedzresnetblur50.bt_in1kzjhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/resnetblur50-84f4748f.pthzresnetblur50d.untrainedzresnetblur101d.untrainedzresnetaa34d.untrainedzresnetaa50.a1h_in1kzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/resnetaa50_a1h-4cf422b3.pthzseresnetaa50d.untrainedzseresnextaa101d_32x8d.ah_in1kzzhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tpu-weights/seresnextaa101d_32x8d_ah-83c8ae12.pthzresnetrs50.tf_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs50_ema-6b53758b.pthrS  rU  gQ?)r[  r2  r7  r8  r9  rJ  r:  r=  zresnetrs101.tf_in1kzuhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs101_i192_ema-1509bbf6.pth)r@      rd  zresnetrs152.tf_in1kzuhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs152_i256_ema-a9aff7f9.pthzresnetrs200.tf_in1kzohttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tpu-weights/resnetrs200_c-6b698b88.pthzresnetrs270.tf_in1kzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs270_ema-b40e674c.pthzresnetrs350.tf_in1kzuhttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs350_i256_ema-5a1aa8f1.pthzresnetrs420.tf_in1kzphttps://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rs-weights/resnetrs420_ema-972dee69.pth)r@     re  zresnet18.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet18_v1b-0757602b.pthzresnet34.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet34_v1b-c6d82d59.pthzresnet50.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet50_v1b-0ebe02e2.pthzresnet101.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet101_v1b-3b017079.pthzresnet152.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet152_v1b-c1edb0dd.pthzresnet50c.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet50_v1c-48092f55.pthzresnet101c.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet101_v1c-1f26822a.pthzresnet152c.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet152_v1c-a3bb0b98.pthzresnet50d.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet50_v1d-818a1b1b.pthzresnet101d.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet101_v1d-0f9c8644.pthzresnet152d.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet152_v1d-bd354e12.pthzresnet50s.gluon_in1kzrhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet50_v1s-1762acc0.pthzresnet101s.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet101_v1s-60fe0cc1.pthzresnet152s.gluon_in1kzshttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnet152_v1s-dcc41b81.pthzresnext50_32x4d.gluon_in1kzuhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext50_32x4d-e6a097c1.pthzresnext101_32x4d.gluon_in1kzvhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext101_32x4d-b253c8c4.pthzresnext101_64x4d.gluon_in1kzvhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext101_64x4d-f9a8e184.pthzseresnext50_32x4d.gluon_in1kzwhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_seresnext50_32x4d-90cf2d6e.pthzseresnext101_32x4d.gluon_in1kzxhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_seresnext101_32x4d-cf52900d.pthzseresnext101_64x4d.gluon_in1kzxhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_seresnext101_64x4d-f9926f93.pthzsenet154.gluon_in1kznhttps://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_senet154-70a1a3c0.pthztest_resnet.r160_in1k)r[  r;  r<  r9  r7  r8  r=  c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-10-T model.
    r   r   r   r   r   deep_tieredTr   r   r   r   r   	resnet10tr   r"   r1  r/  r   
model_argss      r,   rj  rj    4     J|VcnrsJ+zPT*5O5OPPr.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-14-T model.
    rg  r   rh  Tri  	resnet14tr   r#   r1  rl  s      r,   rp  rp    rn  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z"Constructs a ResNet-18 model.
    r)   r)   r)   r)   r   r   resnet18rk  rl  s      r,   ru  ru    ,     J|<J*jOD4Nv4NOOr.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-18-D model.
    rs  r   r   Tri  	resnet18drk  rl  s      r,   rx  rx    4     J|V\gklJ+zPT*5O5OPPr.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z"Constructs a ResNet-34 model.
    r@   rx   rZ  r@   rt  resnet34rk  rl  s      r,   r|  r|    rv  r.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-34-D model.
    r{  r   r   Tri  	resnet34drk  rl  s      r,   r~  r~    ry  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z"Constructs a ResNet-26 model.
    rs  rt  resnet26rq  rl  s      r,   r  r    rv  r.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-26-T model.
    rs  r   rh  Tri  	resnet26trq  rl  s      r,   r  r    rn  r.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-26-D model.
    rs  r   r   Tri  	resnet26drq  rl  s      r,   r  r    ry  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z"Constructs a ResNet-50 model.
    r{  rt  resnet50rq  rl  s      r,   r  r    rv  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-50-C model.
    r{  r   r   r   r   r   r   	resnet50crq  rl  s      r,   r  r    1     J|V\]J+zPT*5O5OPPr.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-50-D model.
    r{  r   r   Tri  	resnet50drq  rl  s      r,   r  r    ry  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-50-S model.
    r{  r0   r   r  	resnet50srq  rl  s      r,   r  r    r  r.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z$Constructs a ResNet-50-T model.
    r{  r   rh  Tri  	resnet50trq  rl  s      r,   r  r    rn  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z#Constructs a ResNet-101 model.
    r@   rx      r@   rt  	resnet101rq  rl  s      r,   r  r    ,     J}=J+zPT*5O5OPPr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-101-C model.
    r  r   r   r  
resnet101crq  rl  s      r,   r  r    1     J}W]^J,
Qd:6P6PQQr.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-101-D model.
    r  r   r   Tri  
resnet101drq  rl  s      r,   r  r    4     J}W]hlmJ,
Qd:6P6PQQr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-101-S model.
    r  r0   r   r  
resnet101srq  rl  s      r,   r  r    r  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z#Constructs a ResNet-152 model.
    r@   r   $   r@   rt  	resnet152rq  rl  s      r,   r  r    r  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-152-C model.
    r  r   r   r  
resnet152crq  rl  s      r,   r  r  '  r  r.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-152-D model.
    r  r   r   Tri  
resnet152drq  rl  s      r,   r  r  /  r  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-152-S model.
    r  r0   r   r  
resnet152srq  rl  s      r,   r  r  7  r  r.   c           	      R    t        t        d      }t        d| fi t        |fi |S )z#Constructs a ResNet-200 model.
    r@      r  r@   rt  	resnet200rq  rl  s      r,   r  r  ?  s,     J~>J+zPT*5O5OPPr.   c           	      X    t        t        dddd      }t        d| fi t        |fi |S )z%Constructs a ResNet-200-D model.
    r  r   r   Tri  
resnet200drq  rl  s      r,   r  r  G  s4     J~"X^imnJ,
Qd:6P6PQQr.   c           	      T    t        t        dd      }t        d| fi t        |fi |S )aO  Constructs a Wide ResNet-50-2 model.
    The model is the same as ResNet except for the bottleneck number of channels
    which is twice larger in every block. The number of channels in outer 1x1
    convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
    channels, and in Wide ResNet-50-2 has 2048-1024-2048.
    r{  r   r   r   r5   wide_resnet50_2rq  rl  s      r,   r  r  O  s/     J|LJ+ZV4
;Uf;UVVr.   c           	      T    t        t        dd      }t        d| fi t        |fi |S )zConstructs a Wide ResNet-101-2 model.
    The model is the same as ResNet except for the bottleneck number of channels
    which is twice larger in every block. The number of channels in outer 1x1
    convolutions is the same.
    r  r   r  wide_resnet101_2rq  rl  s      r,   r  r  [  s/     J}MJ,jWD<Vv<VWWr.   c           	      T    t        t        dd      }t        d| fi t        |fi |S )z.Constructs a ResNet-50 model w/ GroupNorm
    r{  	groupnorm)r   r   r9   resnet50_gnrq  rl  s      r,   r  r  f  s.     J|TJ-RtJ7Q&7QRRr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z(Constructs a ResNeXt50-32x4d model.
    r{  r   rx   r   r   r4   r5   resnext50_32x4drq  rl  s      r,   r  r  n  s2     J|XYZJ+ZV4
;Uf;UVVr.   c           	      \    t        t        dddddd      }t        d| fi t        |fi |S )zVConstructs a ResNeXt50d-32x4d model. ResNext50 w/ deep stem & avg pool downsample
    r{  r   rx   r   T)r   r   r4   r5   r   r   r   resnext50d_32x4drq  rl  s      r,   r  r  v  s=     B1$8J ,jWD<Vv<VWWr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z*Constructs a ResNeXt-101 32x4d model.
    r  r   rx   r  resnext101_32x4drq  rl  s      r,   r  r    2     J}"YZ[J,jWD<Vv<VWWr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z*Constructs a ResNeXt-101 32x8d model.
    r  r   r   r  resnext101_32x8drq  rl  s      r,   r  r    r  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z*Constructs a ResNeXt-101 32x16d model
    r  r   r   r  resnext101_32x16drq  rl  s      r,   r  r    3     J}"Y[\J-zXT*=WPV=WXXr.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z*Constructs a ResNeXt-101 32x32d model
    r  r   r  resnext101_32x32drq  rl  s      r,   r  r    r  r.   c           	      V    t        t        ddd      }t        d| fi t        |fi |S )z)Constructs a ResNeXt101-64x4d model.
    r  r0   rx   r  resnext101_64x4drq  rl  s      r,   r  r    r  r.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	zConstructs an ECA-ResNeXt-26-T model.
    This is technically a 28 layer ResNet, like a 'D' bag-of-tricks model but with tiered 24, 32, 64 channels
    in the deep stem and ECA attn.
    rs  r   rh  Tecar:   r   r   r   r   r   r   ecaresnet26trq  rl  s      r,   r  r    s@     "$45;QSJ .*SZ8R68RSSr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z-Constructs a ResNet-50-D model with eca.
    r{  r   r   Tr  r  r  ecaresnet50drq  rl  s      r,   r  r    s@     "Y]5)+J .*SZ8R68RSSr.   c           
      r    t        t        ddddt        d            }t        d| fd	dit        |fi |S )
zConstructs a ResNet-50-D model pruned with eca.
        The pruning has been obtained using https://arxiv.org/pdf/2002.08258.pdf
    r{  r   r   Tr  r  r  ecaresnet50d_prunedprunedrq  rl  s      r,   r  r    sH    
 "Y]5)+J /gDgDQ[Lf_eLfggr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	zConstructs an ECA-ResNet-50-T model.
    Like a 'D' bag-of-tricks model but with tiered 24, 32, 64 channels in the deep stem and ECA attn.
    r{  r   rh  Tr  r  r  ecaresnet50trq  rl  s      r,   r  r    s@    
 "$45;QSJ .*SZ8R68RSSr.   c           	      l    t        t        dddt        d            }t        d| fi t        |fi |S )z3Constructs a ResNet-50-D light model with eca.
    )r   r      r@   r   Tr  r  )r   r   r   r   r   ecaresnetlightrq  rl  s      r,   r  r    s>     25)+J *JU$z:TV:TUUr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z.Constructs a ResNet-101-D model with eca.
    r  r   r   Tr  r  r  ecaresnet101drq  rl  s      r,   r  r    s@     2Z^5)+J /:Tj9SF9STTr.   c           
      r    t        t        ddddt        d            }t        d| fd	dit        |fi |S )
zConstructs a ResNet-101-D model pruned with eca.
       The pruning has been obtained using https://arxiv.org/pdf/2002.08258.pdf
    r  r   r   Tr  r  r  ecaresnet101d_prunedr  rq  rl  s      r,   r  r    sH    
 2Z^5)+J 0*hThTR\Mg`fMghhr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z.Constructs a ResNet-200-D model with ECA.
    r  r   r   Tr  r  r  ecaresnet200drq  rl  s      r,   r  r    @     B&[_5)+J /:Tj9SF9STTr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z.Constructs a ResNet-269-D model with ECA.
    r@      0   r   r   r   Tr  r  r  ecaresnet269drq  rl  s      r,   r  r    r  r.   c                 r    t        t        ddddddt        d            }t        d	| fi t        |fi |S )
zConstructs an ECA-ResNeXt-26-T model.
    This is technically a 28 layer ResNet, like a 'D' bag-of-tricks model but with tiered 24, 32, 64 channels
    in the deep stem. This model replaces SE module with the ECA module
    rs  r   rx   rh  Tr  r  r   r   r4   r5   r   r   r   r   ecaresnext26t_32x4drq  rl  s      r,   r  r    G     2!XZ$45;QSJ /ZtJ?YRX?YZZr.   c                 r    t        t        ddddddt        d            }t        d	| fi t        |fi |S )
zConstructs an ECA-ResNeXt-50-T model.
    This is technically a 28 layer ResNet, like a 'D' bag-of-tricks model but with tiered 24, 32, 64 channels
    in the deep stem. This model replaces SE module with the ECA module
    rs  r   rx   rh  Tr  r  r  ecaresnext50t_32x4drq  rl  s      r,   r  r    r  r.   c           	      h    t        t        dt        d            }t        d| fi t        |fi |S )Nrs  rS   r  r   r   r   
seresnet18rk  rl  s      r,   r  r    2    J|X\H]^J,
Qd:6P6PQQr.   c           	      h    t        t        dt        d            }t        d| fi t        |fi |S )Nr{  rS   r  r  
seresnet34rk  rl  s      r,   r  r  %  r  r.   c           	      h    t        t        dt        d            }t        d| fi t        |fi |S )Nr{  rS   r  r  
seresnet50rq  rl  s      r,   r  r  +  r  r.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	Nr{  r   rh  TrS   r  r  seresnet50trq  rl  s      r,   r  r  1  s=    2$$"79J -RtJ7Q&7QRRr.   c           	      h    t        t        dt        d            }t        d| fi t        |fi |S )Nr  rS   r  r  seresnet101rq  rl  s      r,   r  r  9  2    J}Y]I^_J-RtJ7Q&7QRRr.   c           	      h    t        t        dt        d            }t        d| fi t        |fi |S )Nr  rS   r  r  seresnet152rq  rl  s      r,   r  r  ?  r  r.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	Nr  r   r   TrS   r  r  seresnet152drq  rl  s      r,   r  r  E  s=    2$$"79J .*SZ8R68RSSr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z2Constructs a ResNet-200-D model with SE attn.
    r  r   r   TrS   r  r  seresnet200drq  rl  s      r,   r  r  M  ?     B&$$"79J .*SZ8R68RSSr.   c           
      n    t        t        ddddt        d            }t        d| fi t        |fi |S )	z2Constructs a ResNet-269-D model with SE attn.
    r  r   r   TrS   r  r  seresnet269drq  rl  s      r,   r  r  W  r  r.   c                 r    t        t        ddddddt        d            }t        d	| fi t        |fi |S )
zConstructs a SE-ResNeXt-26-D model.`
    This is technically a 28 layer ResNet, using the 'D' modifier from Gluon / bag-of-tricks for
    combination of deep stem and avg_pool in downsample.
    rs  r   rx   r   TrS   r  r  seresnext26d_32x4drq  rl  s      r,   r  r  a  sG     2!XZ4DD4IKJ .
Yd:>XQW>XYYr.   c                 r    t        t        ddddddt        d            }t        d	| fi t        |fi |S )
zConstructs a SE-ResNet-26-T model.
    This is technically a 28 layer ResNet, like a 'D' bag-of-tricks model but with tiered 24, 32, 64 channels
    in the deep stem.
    rs  r   rx   rh  TrS   r  r  seresnext26t_32x4drq  rl  s      r,   r  r  m  sG     2!XZ$44;PRJ .
Yd:>XQW>XYYr.   c           	      l    t        t        dddt        d            }t        d| fi t        |fi |S )Nr{  r   rx   rS   r  r   r   r4   r5   r   seresnext50_32x4drq  rl  s      r,   r  r  y  s=    2!4(*J -zXT*=WPV=WXXr.   c           	      l    t        t        dddt        d            }t        d| fi t        |fi |S )Nr  r   rx   rS   r  r  seresnext101_32x4drq  rl  s      r,   r  r    =    B14(*J .
Yd:>XQW>XYYr.   c           	      l    t        t        dddt        d            }t        d| fi t        |fi |S )Nr  r   r   rS   r  r  seresnext101_32x8drq  rl  s      r,   r	  r	    r  r.   c                 r    t        t        ddddddt        d            }t        d	| fi t        |fi |S )
Nr  r   r   r   TrS   r  r  seresnext101d_32x8drq  rl  s      r,   r  r    sD    B1$4(*J /ZtJ?YRX?YZZr.   c           	      l    t        t        dddt        d            }t        d| fi t        |fi |S )Nr  r0   rx   rS   r  r  seresnext101_64x4drq  rl  s      r,   r  r    r  r.   c                 r    t        t        ddddddt        d      	      }t        d
| fi t        |fi |S )Nr  r0   rx   r   r@   r)   rS   r  )r   r   r4   r5   r   r   r   r   senet154rq  rl  s      r,   r  r    sC    B1X^qTT=RTJ *jOD4Nv4NOOr.   c           	      \    t        t        dt              }t        d| fi t        |fi |S )z9Constructs a ResNet-18 model with blur anti-aliasing
    rs  r   r   r;   resnetblur18)r   r"   r   r1  rl  s      r,   r  r    .     J|jQJ.*SZ8R68RSSr.   c           	      \    t        t        dt              }t        d| fi t        |fi |S )z9Constructs a ResNet-50 model with blur anti-aliasing
    r{  r  resnetblur50r   r#   r   r1  rl  s      r,   r  r    r  r.   c           	      b    t        t        dt        ddd      }t        d| fi t        |fi |S )z;Constructs a ResNet-50-D model with blur anti-aliasing
    r{  r   r   Tr   r   r;   r   r   r   resnetblur50dr  rl  s      r,   r  r    s:     
$8J /:Tj9SF9STTr.   c           	      b    t        t        dt        ddd      }t        d| fi t        |fi |S )z<Constructs a ResNet-101-D model with blur anti-aliasing
    r  r   r   Tr  resnetblur101dr  rl  s      r,   r  r    s;     $8J *JU$z:TV:TUUr.   c           	      v    t        t        dt        j                  ddd      }t	        d| fi t        |fi |S )z<Constructs a ResNet-34-D model w/ avgpool anti-aliasing
    r{  r   r   Tr  resnetaa34d)r   r"   rJ   r   r1  rl  s      r,   r  r    s?     RT`fquwJ-RtJ7Q&7QRRr.   c           	      p    t        t        dt        j                        }t	        d| fi t        |fi |S )z<Constructs a ResNet-50 model with avgpool anti-aliasing
    r{  r  
resnetaa50r   r#   rJ   r   r1  rl  s      r,   r  r    s2     J|bllSJ,
Qd:6P6PQQr.   c           	      v    t        t        dt        j                  ddd      }t	        d| fi t        |fi |S )z>Constructs a ResNet-50-D model with avgpool anti-aliasing
    r{  r   r   Tr  resnetaa50dr   rl  s      r,   r"  r"    s>     $8J -RtJ7Q&7QRRr.   c           	      v    t        t        dt        j                  ddd      }t	        d| fi t        |fi |S )z?Constructs a ResNet-101-D model with avgpool anti-aliasing
    r  r   r   Tr  resnetaa101dr   rl  s      r,   r$  r$    s>     $8J .*SZ8R68RSSr.   c                     t        t        dt        j                  dddt        d            }t	        d| fi t        |fi |S )	zAConstructs a SE=ResNet-50-D model with avgpool anti-aliasing
    r{  r   r   TrS   r  )r   r   r;   r   r   r   r   seresnetaa50dr   rl  s      r,   r&  r&    sG     $4SWCXZJ /:Tj9SF9STTr.   c                     t        t        ddddddt        j                  t        d      	      }t	        d	| fi t        |fi |S )
IConstructs a SE=ResNeXt-101-D 32x8d model with avgpool anti-aliasing
    r  r   r   r   TrS   r  	r   r   r4   r5   r   r   r   r;   r   seresnextaa101d_32x8dr   rl  s      r,   r*  r*    sL     B1$4(*J 1:\jA[TZA[\\r.   c                     t        t        ddddddt        j                  t        d      		      }t	        d
| fi t        |fi |S )r(  )r@   r  r  rx   r   r   r0   r   TrS   r  r)  seresnextaa201d_32x8dr   rl  s      r,   r,  r,    sL     RA$4(*J 1:\jA[TZA[\\r.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-50 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   rd_ratior{  r   r   Tr  r   r   r   r   r   r   r   
resnetrs50r   r   r   r#   r1  r/  r   r:   rm  s       r,   r1  r1    sS     $$7J"bf4:#>@J ,
Qd:6P6PQQr.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-101 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  r  r   r   Tr  r0  resnetrs101r2  r3  s       r,   r5  r5     S     $$7J2cg4:#>@J -RtJ7Q&7QRRr.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-152 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  r  r   r   Tr  r0  resnetrs152r2  r3  s       r,   r8  r8  -  r6  r.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-200 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  r  r   r   Tr  r0  resnetrs200r2  r3  s       r,   r:  r:  :  S     $$7JB&dh4:#>@J -RtJ7Q&7QRRr.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-270 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  )rx      5   rx   r   r   Tr  r0  resnetrs270r2  r3  s       r,   r?  r?  G  r;  r.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-350 model.
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  )rx   r  H   rx   r   r   Tr  r0  resnetrs350r2  r3  s       r,   rB  rB  U  r;  r.   c                     t        t        d      d      }t        t        dddddt        |      	      }t	        d
| fi t        |fi |S )zConstructs a ResNet-RS-420 model
    Paper: Revisiting ResNets - https://arxiv.org/abs/2103.07579
    Pretrained weights from https://github.com/tensorflow/tpu/tree/bee9c4f6/models/official/resnet/resnet_rs
    rS   r   r.  )rx   ,   W   rx   r   r   Tr  r0  resnetrs420r2  r3  s       r,   rF  rF  b  r;  r.   c           	      z    t        t        t        t        t        gddddd      }t        d| fi t        |fi |S )z)Constructs a tiny ResNet test model.
    rg  r   r   T)r   r  r  `   )r   r   r   r   r   rE   test_resnet)r   r"   r#   r1  rl  s      r,   rI  rI  o  sE     :z:>|$AQSJ -RtJ7Q&7QRRr.   tv_resnet34tv_resnet50tv_resnet101tv_resnet152tv_resnext50_32x4dig_resnext101_32x8dig_resnext101_32x16dig_resnext101_32x32dig_resnext101_32x48dssl_resnet18ssl_resnet50ssl_resnext50_32x4dssl_resnext101_32x4dssl_resnext101_32x8dssl_resnext101_32x16dswsl_resnet18swsl_resnet50swsl_resnext50_32x4dswsl_resnext101_32x4dswsl_resnext101_32x8dswsl_resnext101_32x16dgluon_resnet18_v1bgluon_resnet34_v1bgluon_resnet50_v1bgluon_resnet101_v1bgluon_resnet152_v1bgluon_resnet50_v1cgluon_resnet101_v1cgluon_resnet152_v1cgluon_resnet50_v1dgluon_resnet101_v1dgluon_resnet152_v1dgluon_resnet50_v1sgluon_resnet101_v1sr  )	gluon_resnet152_v1sgluon_resnext50_32x4dgluon_resnext101_32x4dgluon_resnext101_64x4dgluon_seresnext50_32x4dgluon_seresnext101_32x4dgluon_seresnext101_64x4dgluon_senet154seresnext26tn_32x4d)r   )r   r   NN)r   )r   r   r   Fr   r   r*  )r   )rn   r{   	functoolsr   typingr   r   r   r   r   r	   r
   rs   torch.nnrJ   torch.nn.functional
functionalr%  	timm.datar   r   timm.layersr   r   r   r   r   r   r   r   r   r   r   r   _builderr   	_featuresr   _manipulater   	_registryr   r   r    __all__rq   r-   rr   r"   r#   r   r   r'  r   r,  r+  r   r!   r1  r@  rD  rM  rQ  rV  rX  default_cfgsrj  rp  ru  rx  r|  r~  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r  r	  r  r  r  r  r  r  r  r  r  r"  r$  r&  r*  r,  r1  r5  r8  r:  r?  rB  rF  rI  rk   r*   r.   r,   <module>r     s&     @ @ @     AU U U U * + ' Y Y
0S #  S 
d dNj jb (,04  	
  ! T"))_- YY2 (,04  	
  ! T"))_- YY0j5 j$x/@*A j(  !!# "Q tJ/j1AABCGHQ S/Q  S#XQ  	Q 
 Q  Q  Q  Q  Q  Q  4c299n%&T#s(^(<<=Q hPRYY Pf
GC GT G G	c 	T#s(^ 	Is Id38n I
 tCH~ s d38n  tCH~ s d38n  % s	& A F$Xe	s	&  A F$Xe	s	& As	& As	&$ A%s	&* &y+s	&2 $U/CI&W3s	&8 A9s	&> A?s	&D Es	&L tvMs	&R #E/C%ASs	&X &yYs	&` tvas	&f ugs	&n & C&}C	Ios	&x }ys	&~  D FSR_orts	&F AGs	&L AMs	&R ~@Ss	&X ~@Ys	&^ }_s	&d }es	&j u|~ks	&p z|qs	&v y{ws	&| y{}s	&B y{Cs	&H &yIs	&P $U/}C	&Qs	&Z  A[s	&b  Acs	&j  Aks	&r 6Y7ss	&t %}us	&z  AB{s	&@  ABAs	&F  ABGs	&L 6z&SW=	:Ms	&V %}Ws	&\  AB]s	&b  ABcs	&h  ABis	&n 6z&SW=	:os	&x 68ys	&z 6z&SW=	:{s	&D  ~"@Es	&N G+NPOs	&V G+NPWs	&^ G+NP_s	&f G FMaf+N	Pgs	&p H+NPqs	&x $H FMaf+N	Pys	&B H+NPCs	&J $H FMaf+N	PKs	&T tN+N PUs	&\ N FMaf+N	!P]s	&f O+N!Pgs	&n  O FMaf+N	"Pos	&| F A}s	&H  D!EIs	&N u G HOs	&T u G HUs	&Z v G H[s	&` v A Bas	&f |!gs	&n !&(os	&p  C!Dqs	&z tN+N P{s	&B O+N!PCs	&J O+N!PKs	&R N FMaf+N	!PSs	&\  O FMaf+N	"P]s	&n +DQ+[-]os	&v ,TR+[.]ws	&~ ,TR+[.]s	&F	 ,TR+[.]G	s	&T	 'r+r*tU	s	&\	 'r+r*t]	s	&d	 .tx+r0te	s	&l	 /y+r1tm	s	&t	 /y+r1tu	s	&|	 0z+r2t}	s	&J
 $Ty+r&tK
s	&R
 $Ty+r&tS
s	&Z
 +D+r-t[
s	&b
 ,T A+r.tc
s	&j
 ,T A+r.tk
s	&r
 -d B+r/ts
s	&~
 V|&M	;
s	&H zM!;Is	&P exD-QQs	&X $UzD-&QYs	&` U|&M	;as	&j E Dks	&r E Dss	&z F D{s	&B uyD- QCs	&J %e{D-'QKs	&R vQW YSs	&V f B(UY=	:Ws	&d $Ui%@es	&f $Ui%@gs	&l FHms	&n FHos	&p % Bqs	&x % Bys	&@ & BAs	&H 6}Is	&N VOs	&R VXSs	&T VXUs	&V V|&SW=	Ws	&b f&Jcs	&f f&Jgs	&n !&~#os	&v !&#ws	&~ "6 C$Ds	&D #FHEs	&F !% F#GGs	&L "5 G$Ms	&X #Ft3%@Ys	&^ $Vt3&@_s	&d 1&-6S`ps3es	&l -fC/1ms	&r 1$IVcnq3sss	&x %dy-8Ubru'wys	&B Fi$cSCs	&H 6i$cSIs	&N Vi$cSOs	&T %fi$c'SUs	&\ fh]s	&^ Fxz_s	&d v;es	&f ) <gs	&h Vy9is	&j 5~@ks	&r v;ss	&t $U I&us	&@ $} FTS`I	7As	&J 4 D FTS`I	7Ks	&T 4 D FSR_I	7Us	&^ 4} FSR_I	7_s	&h 4~ FSR_I	7is	&r 4 D FSR_I	7ss	&| 4~ HsTaI	7}s	&J 5 ABKs	&P 5 ABQs	&V 5 ABWs	&\ E BC]s	&b E BCcs	&h E Ais	&p U Bqs	&x U Bys	&@ E AAs	&H U BIs	&P U BQs	&X E AYs	&` U Bas	&h U Bis	&p !% D#Eqs	&v "5 E$Fws	&| "5 E$F}s	&B #E F%GCs	&H $U G&HIs	&N $U G&HOs	&T 5|Us	&^ T/D FyJ_s	& s	l Q$ QV Q Q Q$ QV Q Q P PF P P Q$ QV Q Q P PF P P Q$ QV Q Q P PF P P Q$ QV Q Q Q$ QV Q Q P PF P P Q$ QV Q Q Q$ QV Q Q Q$ QV Q Q Q$ QV Q Q Q$ QV Q Q R4 Rf R R R4 Rf R R R4 Rf R R Q$ QV Q Q R4 Rf R R R4 Rf R R R4 Rf R R Q$ QV Q Q R4 Rf R R W W6 W W X XF X X SD Sv S S W W6 W W X XF X X X XF X X X XF X X Y$ YV Y Y Y$ YV Y Y X XF X X TT T T T TT T T T hD hv h h TT T T T Vt V& V V Ud U U U iT i i i Ud U U U Ud U U U [D [v [ [ [D [v [ [ R4 Rf R R
 R4 Rf R R
 R4 Rf R R
 SD Sv S S SD Sv S S
 SD Sv S S
 TT T T T TT T T T TT T T T Z4 Zf Z Z Z4 Zf Z Z Y$ YV Y Y Z4 Zf Z Z Z4 Zf Z Z [D [v [ [ Z4 Zf Z Z P PF P P TT T T T TT T T T Ud U U U Vt V& V V SD Sv S S R4 Rf R R SD Sv S S TT T T T Ud U U U ]d ] ] ] ]d ] ] 	R4 	Rf 	R 	R 	SD 	Sv 	S 	S 	SD 	Sv 	S 	S 	SD 	Sv 	S 	S 	SD 	Sv 	S 	S 	SD 	Sv 	S 	S 	SD 	Sv 	S 	S SD Sv S S H ,'%,'%,' ',' '	,'
 4,' A,' B,' B,' B,' 6,' 6,' D,' F,' F,' H,'  4!,'" 4#,'$ B%,'& D','( D),'* F+,', /-,'. //,'0 /1,'2 13,'4 15,'6 07,'8 29,': 2;,'< 0=,'> 2?,'@ 2A,'B 0C,'D 2E,'F 39;;= ? ?+/W,' ,r.   