§
    ·¯¸j&Ü  ã                   ó  — 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	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ZmZ ddlmZmZmZmZmZmZmZmZ ddlmZm Z m!Z!m"Z" d	d
l#m$Z$ d	dl%m&Z& d	dl'm(Z(m)Z) d	dl*m+Z+m,Z,m-Z- ddgZ.e G d„ d¦  «        ¦   «         Z/de0de1de1fd„Z2	 �dvdee1         dee0         dee1         de0deee1         ee1         f         f
d„Z3	 �dwde0de1de0de1d e1d!e1deee1         e1ee1         f         fd"„Z4	 	 	 	 	 	 	 �dxd$e1d%e1d&e1d'e1d(e1d)eeej5                          d*e6dej5        fd+„Z7	 	 	 	 	 	 	 �dxd$e1d%e1d&e1d'e1d(e1d)eeej5                          d*e6dej8        fd,„Z9	 	 	 	 	 �dyd.ee:         d$e1d%e1d&e1d'e1d(ee1e1f         d)eeej5                          d*e6deej5                 fd/„Z; G d0„ d1ej5        ¦  «        Z< G d2„ d3ej5        ¦  «        Z= G d4„ d5ej5        ¦  «        Z> G d6„ dej5        ¦  «        Z?�dzd8ej5        d9e:d:e6ddfd;„Z@d<e
e:ef         de
e:ef         fd=„ZA eB�d{i d> e/d?d@dAddB¬C¦  «        “dD e/d?dEdFdGdH¬C¦  «        “dI e/d?dEdFdGdHdJ¬K¦  «        “dL e/dMdNdOd?dG¬C¦  «        “dP e/dQdRdSdGdG¬C¦  «        “dT e/dUdVdWd?dX¬C¦  «        “dY e/dZd[dWdMd\¬C¦  «        “d] e/d^d_d`dadb¬C¦  «        “dc e/dddedfdQdg¬C¦  «        “dh e/dUdidjdkdb¬C¦  «        “dl e/dmdndodpdq¬C¦  «        “dr e/dsdtdudvdH¬C¦  «        “dw e/dxdydzdmdb¬C¦  «        “d{ e/d?d@dAddBd|¬}¦  «        “d~ e/dMdd€ddGd|¬}¦  «        “d� e/dMd‚dƒdGd„d|¬}¦  «        “d… e/dQd†d‡dGdˆd|¬}¦  «        “d‰ e/dQd†d‡dGdˆd|dJ¬Š¦  «        “d‹ e/dMdŒd�d?dŽd|¬}¦  «        “d� e/dUd�d‘d?d’d|¬}¦  «        “d“ e/d^d”dOd•dHd|¬}¦  «        “d– e/dpd—d˜d™d\d|¬}¦  «        “dš e/d›dœd�dQdgd|¬}¦  «        “dž e/d›dœd�dQdgd|dJ¬Š¦  «        “dŸ e/dmdndodpdqd|¬}¦  «        “d  e/d¡d¢d£dpdXd|¬}¦  «        “d¤ e/d¥d¦d§d¥d¨d|¬}¦  «        “d© e/dªd«d‡d¬d¨d|¬}¦  «        “d­ e/d®d¯d°d±dŽd|¬}¦  «        “d² e/d³d´d§dµdŽd|¬}¦  «        “d¶ e/d^d”dOd•dHd|d· eedG¬¸¦  «        ¬¹¦  «        “dº e/dHd^d”dOd•d|d»d·¬¼¦  «        “d½ e/d\dpd—d˜d™d|d»d·d¾¬¿¦	  «	        “dÀ e/d’dGdÁdÂdÃdÄd|dd»dÅd·¬Æ¦  «        “dÇ e/dÈdMdÉdÊddÄd|dd»dd·¬Æ¦  «        “dË e/dÈdMdÉdÊddÄd|dd»dÌd·¬Æ¦  «        “ŽZCdÍe:dÎe6de?fdÏ„ZD�d|dÐe:de
e:ef         fdÑ„ZE�d|dÐe:de
e:ef         fdÒ„ZF�d|dÐe:de
e:ef         fdÓ„ZG e+i dÔ eEdÕdÖ¬×¦  «        “dØ eEdÕdÙ¬×¦  «        “dÚ eEdÕdÛ¬×¦  «        “dÜ eEdÕdÝ¬×¦  «        “dÞ eEdÕ¬ß¦  «        “dà eEdÕ¬ß¦  «        “dá eEdÕ¬ß¦  «        “dâ eEdÕdã¬ä¦  «        “då eEdÕdã¬ä¦  «        “dæ eEd7¬ç¦  «        “dè eEdÕdédê¬ë¦  «        “dì eEdÕdídê¬ë¦  «        “dî eEd7¬ç¦  «        “dï eEdÕdðdñdòdódô¬õ¦  «        “dö eEdÕd÷dñdòdódô¬õ¦  «        “dø eEdÕdù¬×¦  «        “dú eGdÕdû¬×¦  «        “i dü eGdÕdý¬×¦  «        “dþ eGdÕdÿ¬×¦  «        “�d  eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d	¬×¦  «        “�d
 eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d¬×¦  «        “�d eGdÕ�d�d�d�ddó�¬¦  «        “�d eGdÕ�d�d�d�ddó�¬¦  «        “�d eGdÕ�d�d�d�ddó�¬¦  «        “�d  eGdÕ�d!�d�¬"¦  «        “¥i �d# eGdÕ�d$�d�¬"¦  «        “�d% eGdÕ�d&�d�¬"¦  «        “�d' eGdÕ�d(�d)�d*�d�ddó�¬+¦  «        “�d, eGdÕ�d(�d)�d-�d�ddó�¬+¦  «        “�d. eGdÕ�d(�d)�d/�d�ddó�¬+¦  «        “�d0 eGdÕ�d(�d)�d1�d�ddó�¬+¦  «        “�d2 eGdÕ�d3d�d(�d)�¬4¦  «        “�d5 eGdÕ�d6d�d(�d)�¬4¦  «        “�d7 eGdÕ�d8d�d(�d)�¬4¦  «        “�d9 eFdÕ¬ß¦  «        “�d: eFdÕ¬ß¦  «        “�d; eFdÕ¬ß¦  «        “�d< eFdÕ¬ß¦  «        “�d= eFdÕ¬ß¦  «        “�d> eFdÕ¬ß¦  «        “�d? eFdÕ¬ß¦  «        “�d@ eFdÕ¬ß¦  «        “¥i �dA eFdÕ¬ß¦  «        “�dB eFdÕ¬ß¦  «        “�dC eFdÕ¬ß¦  «        “�dD eFdÕ¬ß¦  «        “�dE eFdÕ¬ß¦  «        “�dF eFdÕ¬ß¦  «        “�dG eFdÕ¬ß¦  «        “�dH eFdÕ¬ß¦  «        “�dI eFdÕ¬ß¦  «        “�dJ eFdÕ¬ß¦  «        “�dK eFdÕ¬ß¦  «        “�dL eFdÕ¬ß¦  «        “�dM eFdÕ¬ß¦  «        “�dN eFdÕ¬ß¦  «        “�dO eFdÕ¬ß¦  «        “�dP eFdÕ¬ß¦  «        “¥¦  «        ZHe,�d}dÎe6de?f�dQ„¦   «         ZIe,�d}dÎe6de?f�dR„¦   «         ZJe,�d}dÎe6de?f�dS„¦   «         ZKe,�d}dÎe6de?f�dT„¦   «         ZLe,�d}dÎe6de?f�dU„¦   «         ZMe,�d}dÎe6de?f�dV„¦   «         ZNe,�d}dÎe6de?f�dW„¦   «         ZOe,�d}dÎe6de?f�dX„¦   «         ZPe,�d}dÎe6de?f�dY„¦   «         ZQe,�d}dÎe6de?f�dZ„¦   «         ZRe,�d}dÎe6de?f�d[„¦   «         ZSe,�d}dÎe6de?f�d\„¦   «         ZTe,�d}dÎe6de?f�d]„¦   «         ZUe,�d}dÎe6de?f�d^„¦   «         ZVe,�d}dÎe6de?f�d_„¦   «         ZWe,�d}dÎe6de?f�d`„¦   «         ZXe,�d}dÎe6de?f�da„¦   «         ZYe,�d}dÎe6de?f�db„¦   «         ZZe,�d}dÎe6de?f�dc„¦   «         Z[e,�d}dÎe6de?f�dd„¦   «         Z\e,�d}dÎe6de?f�de„¦   «         Z]e,�d}dÎe6de?f�df„¦   «         Z^e,�d}dÎe6de?f�dg„¦   «         Z_e,�d}dÎe6de?f�dh„¦   «         Z`e,�d}dÎe6de?f�di„¦   «         Zae,�d}dÎe6de?f�dj„¦   «         Zbe,�d}dÎe6de?f�dk„¦   «         Zce,�d}dÎe6de?f�dl„¦   «         Zde,�d}dÎe6de?f�dm„¦   «         Zee,�d}dÎe6de?f�dn„¦   «         Zfe,�d}dÎe6de?f�do„¦   «         Zge,�d}dÎe6de?f�dp„¦   «         Zhe,�d}dÎe6de?f�dq„¦   «         Zie,�d}dÎe6de?f�dr„¦   «         Zje,�d}dÎe6de?f�ds„¦   «         Zke,�d}dÎe6de?f�dt„¦   «         Zl e-em�dudËi¦  «         dS (~  a  RegNet X, Y, Z, and more

Paper: `Designing Network Design Spaces` - https://arxiv.org/abs/2003.13678
Original Impl: https://github.com/facebookresearch/pycls/blob/master/pycls/models/regnet.py

Paper: `Fast and Accurate Model Scaling` - https://arxiv.org/abs/2103.06877
Original Impl: None

Based on original PyTorch impl linked above, but re-wrote to use my own blocks (adapted from ResNet here)
and cleaned up with more descriptive variable names.

Weights from original pycls impl have been modified:
* first layer from BGR -> RGB as most PyTorch models are
* removed training specific dict entries from checkpoints and keep model state_dict only
* remap names to match the ones here

Supports weight loading from torchvision and classy-vision (incl VISSL SEER)

A number of custom timm model definitions additions including:
* stochastic depth, gradient checkpointing, layer-decay, configurable dilation
* a pre-activation 'V' variant
* only known RegNet-Z model definitions with pretrained weights

Hacked together by / Copyright 2020 Ross Wightman
é    N)Ú	dataclassÚreplace)Úpartial)ÚAnyÚCallableÚDictÚListÚOptionalÚUnionÚTupleÚType©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚClassifierHeadÚAvgPool2dSameÚConvNormActÚSEModuleÚDropPathÚGroupNormActÚcalculate_drop_path_ratesÚget_device_dtype)Úget_act_layerÚget_norm_act_layerÚcreate_conv2dÚmake_divisibleé   )Úbuild_model_with_cfg)Úfeature_take_indices)Úcheckpoint_seqÚnamed_apply)Úgenerate_default_cfgsÚregister_modelÚregister_model_deprecationsÚRegNetÚ	RegNetCfgc                   ó  — e Zd ZU dZdZeed<   dZeed<   dZe	ed<   dZ
e	ed	<   d
Zeed<   dZe	ed<   dZe	ed<   dZe	ed<   dZeed<   dZee         ed<   dZeed<   dZeed<   dZeed<   dZeeef         ed<   dZeeef         ed<   dS )r&   z"RegNet architecture configuration.é   ÚdepthéP   Úw0çq=
×£PE@ÚwaçHáz®G@Úwmé   Ú
group_sizeç      ð?Úbottle_ratioç        Úse_ratioÚgroup_min_ratioé    Ú
stem_widthÚconv1x1Ú
downsampleFÚ
linear_outÚpreactr   Únum_featuresÚreluÚ	act_layerÚ	batchnormÚ
norm_layerN)Ú__name__Ú
__module__Ú__qualname__Ú__doc__r)   ÚintÚ__annotations__r+   r-   Úfloatr/   r1   r3   r5   r6   r8   r:   r
   Ústrr;   Úboolr<   r=   r?   r   r   rA   © ó    úb/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/regnet.pyr&   r&   -   s  € € € € € € à,Ð,Ø€Eˆ3€O€O�OØ€Bˆ€L€L�LØ€BˆÐÐÑØ€BˆÐÐÑØ€J�ÐÐÑØ€L�%ÐÐÑØ€HˆeÐÐÑØ€O�UÐÐÑØ€J�ÐÐÑØ )€J�˜”Ð)Ð)Ñ)Ø€J�ÐÐÑØ€FˆDÐÐÑØ€L�#ÐÐÑØ&,€Iˆu�S˜(�]Ô#Ð,Ð,Ñ,Ø'2€J��c˜8�mÔ$Ð2Ð2Ñ2Ð2Ð2rL   ÚfÚqÚreturnc                 óF   — t          t          | |z  ¦  «        |z  ¦  «        S )zºConverts a float to the closest non-zero int divisible by q.

    Args:
        f: Input float value.
        q: Quantization divisor.

    Returns:
        Quantized integer value.
    )rF   Úround)rN   rO   s     rM   Úquantize_floatrS   A   s!   € õ �u�Q˜‘U‰|Œ|˜aÑÑ Ô Ð rL   r4   ÚwidthsÚbottle_ratiosÚgroupsÚ	min_ratioc                 ó  ‡— d„ t          | |¦  «        D ¦   «         }d„ t          ||¦  «        D ¦   «         }‰rˆfd„t          ||¦  «        D ¦   «         }nd„ t          ||¦  «        D ¦   «         }d„ t          ||¦  «        D ¦   «         } | |fS )a,  Adjusts the compatibility of widths and groups.

    Args:
        widths: List of channel widths.
        bottle_ratios: List of bottleneck ratios.
        groups: List of group sizes.
        min_ratio: Minimum ratio for divisibility.

    Returns:
        Tuple of adjusted widths and groups.
    c                 ó8   — g | ]\  }}t          ||z  ¦  «        ‘ŒS rK   ©rF   )Ú.0ÚwÚbs      rM   ú
<listcomp>z-adjust_widths_groups_comp.<locals>.<listcomp>_   s&   € ÐKÐKÐK©¨¨1�˜Q ™U™œÐKÐKÐKrL   c                 ó4   — g | ]\  }}t          ||¦  «        ‘ŒS rK   )Úmin)r[   ÚgÚw_bots      rM   r^   z-adjust_widths_groups_comp.<locals>.<listcomp>`   s$   € ÐKÐKÐK¡  5�c�!�U‰mŒmÐKÐKÐKrL   c                 ó8   •— g | ]\  }}t          ||‰¦  «        ‘ŒS rK   )r   )r[   rb   ra   rW   s      €rM   r^   z-adjust_widths_groups_comp.<locals>.<listcomp>c   s)   ø€ ÐpÐpÐpÁXÀUÈA�^¨E°1°iÑ@Ô@ÐpÐpÐprL   c                 ó4   — g | ]\  }}t          ||¦  «        ‘ŒS rK   )rS   )r[   rb   ra   s      rM   r^   z-adjust_widths_groups_comp.<locals>.<listcomp>e   s&   € ÐeÐeÐe¹(¸%À�^¨E°1Ñ5Ô5ÐeÐeÐerL   c                 ó8   — g | ]\  }}t          ||z  ¦  «        ‘ŒS rK   rZ   )r[   rb   r]   s      rM   r^   z-adjust_widths_groups_comp.<locals>.<listcomp>f   s&   € ÐSÐSÐS¡ ¨�c�%˜!‘)‰nŒnÐSÐSÐSrL   )Úzip)rT   rU   rV   rW   Úbottleneck_widthss      ` rM   Úadjust_widths_groups_comprh   N   s½   ø€ ð" LÐKµ°F¸MÑ0JÔ0JÐKÑKÔKÐØKÐK­C°Ð8IÑ,JÔ,JÐKÑKÔK€FØð fàpÐpÐpÐpÕQTÐUfÐhnÑQoÔQoÐpÑpÔpÐÐàeÐeÅcÐJ[Ð]cÑFdÔFdÐeÑeÔeÐØSÐS­SÐ1BÀMÑ-RÔ-RÐSÑSÔS€FØ�6ˆ>ÐrL   é   Úwidth_slopeÚwidth_initialÚ
width_multr)   r1   Úquantc                 ó   ‡— | dk    r|dk    r|dk    r	||z  dk    sJ ‚t          j        |dt           j        ¬¦  «        | z  |z   }t          j        t          j        ||z  ¦  «        t          j        |¦  «        z  ¦  «        }t          j        |t          j        ||¦  «        z  |z  ¦  «        |z  }t          t          j        |¦  «        ¦  «        t          | 
                    ¦   «                              ¦   «         ¦  «        dz   }
}	t          j        ˆfd„t          |	¦  «        D ¦   «         dt           j        ¬¦  «        }| 	                    ¦   «                              ¦   «         |	|                     ¦   «         fS )au  Generates per block widths from RegNet parameters.

    Args:
        width_slope: Slope parameter for width progression.
        width_initial: Initial width.
        width_mult: Width multiplier.
        depth: Network depth.
        group_size: Group convolution size.
        quant: Quantization factor.

    Returns:
        Tuple of (widths, num_stages, groups).
    r   r   Úcpu©ÚdeviceÚdtypec                 ó   •— g | ]}‰‘ŒS rK   rK   )r[   Ú_r1   s     €rM   r^   z#generate_regnet.<locals>.<listcomp>‡   s   ø€ ÐAÐAÐA¨!˜:ÐAÐAÐArL   )ÚtorchÚarangeÚfloat32rR   ÚlogÚmathÚpowÚlenÚuniquerF   ÚmaxÚitemÚtensorÚrangeÚint32Útolist)rj   rk   rl   r)   r1   rm   Úwidths_contÚ
width_expsrT   Ú
num_stagesÚ	max_stagerV   s       `       rM   Úgenerate_regnetr‡   j   sU  ø€ ð* ˜!ÒÐ °Ò 1Ð 1°jÀ1²n°nÈÐY^ÑI^ÐbcÒIcÐIcÐIcÐcõ ”,˜u¨U½%¼-ÐHÑHÔHÈ;ÑVÐYfÑf€KÝ”�UœY {°]Ñ'BÑCÔCÅdÄhÈzÑFZÔFZÑZÑ[Ô[€JÝŒ[˜-­%¬)°JÀ
Ñ*KÔ*KÑKÈuÑTÑUÔUÐX]Ñ]€FÝ¥¤¨VÑ 4Ô 4Ñ5Ô5µs¸:¿>º>Ñ;KÔ;K×;PÒ;PÑ;RÔ;RÑ7SÔ7SÐVWÑ7W�	€JÝŒ\ÐAÐAÐAÐA­u°ZÑ/@Ô/@ÐAÑAÔAÈ%ÕW\ÔWbÐcÑcÔc€FØ�:Š:‰<Œ<×ÒÑ Ô  *¨f¯mªm©o¬oÐ=Ð=rL   FÚin_chsÚout_chsÚkernel_sizeÚstrideÚdilationrA   r<   c	                 ó¬   — ||dœ}	|pt           j        }|dk    r|dk    rdn|}|dk    r|nd}|rt          | ||f||dœ|	¤ŽS t          | ||f|||ddœ|	¤ŽS )am  Create convolutional downsampling module.

    Args:
        in_chs: Input channels.
        out_chs: Output channels.
        kernel_size: Convolution kernel size.
        stride: Convolution stride.
        dilation: Convolution dilation.
        norm_layer: Normalization layer.
        preact: Use pre-activation.

    Returns:
        Downsampling module.
    rp   r   )r‹   rŒ   F)r‹   rŒ   rA   Ú	apply_act)ÚnnÚBatchNorm2dr   r   )
rˆ   r‰   rŠ   r‹   rŒ   rA   r<   rq   rr   Údds
             rM   Údownsample_convr’   ‹   sÅ   € ð2  UÐ	+Ð	+€BØÐ-�rœ~€JØ ’{�{ x°1¢} }�!�!¸+€KØ&¨š?˜?ˆxˆx°€HØð 
ÝØØØð
ð Øð
ð 
ð ð
ð 
ð 	
õ ØØØð	
ð ØØ!Øð	
ð 	
ð ð	
ð 	
ð 		
rL   c	                 óD  — ||dœ}	|pt           j        }|dk    r|nd}
t          j        ¦   «         }|dk    s|dk    r.|
dk    r|dk    rt          nt           j        } |d|
dd¬¦  «        }|rt          | |dfddi|	¤Ž}nt          | |dfd|ddœ|	¤Ž}t          j        ||gŽ S )	aà  Create average pool downsampling module.

    AvgPool Downsampling as in 'D' ResNet variants. This is not in RegNet space but I might experiment.

    Args:
        in_chs: Input channels.
        out_chs: Output channels.
        kernel_size: Convolution kernel size.
        stride: Convolution stride.
        dilation: Convolution dilation.
        norm_layer: Normalization layer.
        preact: Use pre-activation.

    Returns:
        Sequential downsampling module.
    rp   r   é   TF)Ú	ceil_modeÚcount_include_padr‹   )r‹   rA   rŽ   )r�   r�   ÚIdentityr   Ú	AvgPool2dr   r   Ú
Sequential)rˆ   r‰   rŠ   r‹   rŒ   rA   r<   rq   rr   r‘   Ú
avg_strideÚpoolÚavg_pool_fnÚconvs                 rM   Údownsample_avgrž   ¾   sÛ   € ð6  UÐ	+Ð	+€BØÐ-�rœ~€JØ# qš=˜=��¨a€JÝŒ;‰=Œ=€DØ�‚z€z�X ’\�\Ø'1°Q¢ ¸8Àaº<¸<•m�mÍRÌ\ˆØˆ{˜1˜j°DÈEÐRÑRÔRˆØð gÝ˜V W¨aÐ@Ð@¸Ð@¸RÐ@Ð@ˆˆå˜6 7¨AÐf°aÀJÐZ_ÐfÐfÐceÐfÐfˆÝŒ=˜4 ˜,Ð'Ð'rL   ©r   r   Údownsample_typec
                 óø   — ||	dœ}
| dv sJ ‚||k    s|dk    s|d         |d         k    r?t          d	||d         ||dœ|
¤Ž}| sdS | dk    rt          ||fi |¤ŽS t          ||fd|i|¤ŽS t          j        ¦   «         S )
aË  Create shortcut connection for residual blocks.

    Args:
        downsample_type: Type of downsampling ('avg', 'conv1x1', or None).
        in_chs: Input channels.
        out_chs: Output channels.
        kernel_size: Kernel size for conv downsampling.
        stride: Stride for downsampling.
        dilation: Dilation rates.
        norm_layer: Normalization layer.
        preact: Use pre-activation.

    Returns:
        Shortcut module or None.
    rp   )Úavgr9   Ú Nr   r   )r‹   rŒ   rA   r<   Nr¢   rŠ   rK   )Údictrž   r’   r�   r—   )r    rˆ   r‰   rŠ   r‹   rŒ   rA   r<   rq   rr   r‘   Údargss               rM   Úcreate_shortcutr¦   ç   sÆ   € ð6  UÐ	+Ð	+€BØÐ:Ð:Ð:Ð:Ð:Ø�ÒÐ˜F ašK˜K¨8°A¬;¸(À1¼+Ò+EÐ+EÝÐe˜F¨X°a¬[ÀZÐX^ÐeÐeÐbdÐeÐeˆØð 	VØ�4Ø Ò%Ð%Ý! &¨'Ð;Ð;°UÐ;Ð;Ð;å" 6¨7ÐUÐUÀÐUÈuÐUÐUÐUåŒ{‰}Œ}ÐrL   c                   ó  ‡ — e 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ef         de	dede	de
dedeej                 deej                 deeej                          de	fˆ fd„Zdd„Zdej        dej        fd„Zˆ xZS )Ú
BottleneckzúRegNet Bottleneck block.

    This is almost exactly the same as a ResNet Bottleneck. The main difference is the SE block is moved from
    after conv3 to after conv2. Otherwise, it's just redefining the arguments for groups/bottleneck channels.
    r   rŸ   ç      Ð?r9   FNr4   rˆ   r‰   r‹   rŒ   r3   r1   r5   r:   r;   r?   rA   Ú
drop_blockÚdrop_path_ratec           	      óì  •— ||dœ}t          ¦   «                              ¦   «          t          |
¦  «        }
t          t	          ||z  ¦  «        ¦  «        }||z  }t          |
|¬¦  «        }t          ||fddi|¤|¤Ž| _        t          ||fd||d         ||dœ|¤|¤Ž| _        |r5t          t	          ||z  ¦  «        ¦  «        }t          |f||
dœ|¤Ž| _
        nt          j        ¦   «         | _
        t          ||fdd	d
œ|¤|¤Ž| _        |	rt          j        ¦   «         n	 |
¦   «         | _        t          |||fd|||dœ|¤Ž| _        |dk    rt#          |¦  «        nt          j        ¦   «         | _        dS )a³  Initialize RegNet Bottleneck block.

        Args:
            in_chs: Input channels.
            out_chs: Output channels.
            stride: Convolution stride.
            dilation: Dilation rates for conv2 and shortcut.
            bottle_ratio: Bottleneck ratio (reduction factor).
            group_size: Group convolution size.
            se_ratio: Squeeze-and-excitation ratio.
            downsample: Shortcut downsampling type.
            linear_out: Use linear activation for output.
            act_layer: Activation layer.
            norm_layer: Normalization layer.
            drop_block: Drop block layer.
            drop_path_rate: Stochastic depth drop rate.
        rp   ©r?   rA   rŠ   r   é   r   )rŠ   r‹   rŒ   rV   Ú
drop_layer©Úrd_channelsr?   F)rŠ   rŽ   )rŠ   r‹   rŒ   rA   N)ÚsuperÚ__init__r   rF   rR   r¤   r   Úconv1Úconv2r   Úser�   r—   Úconv3Úact3r¦   r:   r   Ú	drop_path)Úselfrˆ   r‰   r‹   rŒ   r3   r1   r5   r:   r;   r?   rA   rª   r«   rq   rr   r‘   Úbottleneck_chsrV   ÚcargsÚse_channelsÚ	__class__s                        €rM   r³   zBottleneck.__init__  sØ  ø€ ðF ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐÝ! )Ñ,Ô,ˆ	Ý�U 7¨\Ñ#9Ñ:Ô:Ñ;Ô;ˆØ :Ñ-ˆå˜y°ZÐ@Ñ@Ô@ˆÝ  ¨ÐVÐVÀQÐVÈ%ÐVÐSUÐVÐVˆŒ
Ý ØØð

ð ØØ˜a”[ØØ!ð

ð 

ð ð

ð ð

ð 

ˆŒ
ð ð 	$Ý�e F¨XÑ$5Ñ6Ô6Ñ7Ô7ˆKÝ˜~Ðb¸;ÐR[ÐbÐbÐ_aÐbÐbˆDŒGˆGå”k‘m”mˆDŒGÝ  °ÐhÀaÐSXÐhÐhÐ\aÐhÐegÐhÐhˆŒ
Ø%/Ð@•B”K‘M”M�M°Y°Y±[´[ˆŒ	Ý)ØØØð	
ð ØØØ!ð	
ð 	
ð ð	
ð 	
ˆŒð 6DÀaÒ5GÐ5G� .Ñ1Ô1Ð1ÍRÌ[É]Ì]ˆŒˆˆrL   rP   c                 ób   — t           j                             | j        j        j        ¦  «         dS )z1Zero-initialize the last batch norm in the block.N)r�   ÚinitÚzeros_r·   ÚbnÚweight©rº   s    rM   Úzero_init_lastzBottleneck.zero_init_last`  s#   € å
Œ�Š�t”z”}Ô+Ñ,Ô,Ð,Ð,Ð,rL   Úxc                 ó@  — |}|                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        �+|                      |¦  «        |                      |¦  «        z   }|                      |¦  «        }|S ©zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        )r´   rµ   r¶   r·   r:   r¹   r¸   ©rº   rÆ   Úshortcuts      rM   ÚforwardzBottleneck.forwardd  s…   € ð ˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØ�GŠG�A‰JŒJˆØ�JŠJ�q‰MŒMˆØŒ?Ð&ð —’˜qÑ!Ô! D§O¢O°HÑ$=Ô$=Ñ=ˆAØ�IŠI�a‰LŒLˆØˆrL   ©rP   N©rB   rC   rD   rE   r�   ÚReLUr�   rF   r   rH   rI   rJ   r   ÚModuler
   r³   rÅ   ru   ÚTensorrË   Ú__classcell__©r¾   s   @rM   r¨   r¨     s{  ø€ € € € € ðð ð Ø(.Ø"#ØØ"Ø'Ø$Ø)+¬Ø*,¬.Ø48Ø$&ØØð!G[ð G[àðG[ð ðG[ð ð	G[ð
 ˜C ˜H”oðG[ð  ðG[ð ðG[ð ðG[ð ðG[ð ðG[ð ˜BœI”ðG[ð ˜RœYœðG[ð !  b¤i¤Ô1ðG[ð "ðG[ð G[ð G[ð G[ð G[ð G[ðR-ð -ð -ð -ð˜œð ¨%¬,ð ð ð ð ð ð ð ð rL   r¨   c                   ó  ‡ — e 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ef         de	dede	de
dedeej                 deej                 deeej                          de	fˆ fd„Zdd„Zdej        dej        fd„Zˆ xZS )ÚPreBottleneckznPre-activation RegNet Bottleneck block.

    Similar to Bottleneck but with pre-activation normalization.
    r   rŸ   r©   r9   FNr4   rˆ   r‰   r‹   rŒ   r3   r1   r5   r:   r;   r?   rA   rª   r«   c                 óÈ  •— ||dœ}t          ¦   «                              ¦   «          t          ||
¦  «        }t          t	          ||z  ¦  «        ¦  «        }||z  } ||fi |¤Ž| _        t          ||fddi|¤Ž| _         ||fi |¤Ž| _        t          ||fd||d         |dœ|¤Ž| _	        |r5t          t	          ||z  ¦  «        ¦  «        }t          |f||
dœ|¤Ž| _        nt          j        ¦   «         | _         ||fi |¤Ž| _        t          ||fddi|¤Ž| _        t!          |||fd||dd	œ|¤Ž| _        |dk    rt%          |¦  «        nt          j        ¦   «         | _        d
S )aÂ  Initialize pre-activation RegNet Bottleneck block.

        Args:
            in_chs: Input channels.
            out_chs: Output channels.
            stride: Convolution stride.
            dilation: Dilation rates for conv2 and shortcut.
            bottle_ratio: Bottleneck ratio (reduction factor).
            group_size: Group convolution size.
            se_ratio: Squeeze-and-excitation ratio.
            downsample: Shortcut downsampling type.
            linear_out: Use linear activation for output.
            act_layer: Activation layer.
            norm_layer: Normalization layer.
            drop_block: Drop block layer.
            drop_path_rate: Stochastic depth drop rate.
        rp   rŠ   r   r®   r   )rŠ   r‹   rŒ   rV   r°   T)rŠ   r‹   rŒ   r<   N)r²   r³   r   rF   rR   Únorm1r   r´   Únorm2rµ   r   r¶   r�   r—   Únorm3r·   r¦   r:   r   r¹   )rº   rˆ   r‰   r‹   rŒ   r3   r1   r5   r:   r;   r?   rA   rª   r«   rq   rr   r‘   Únorm_act_layerr»   rV   r½   r¾   s                        €rM   r³   zPreBottleneck.__init__€  sÚ  ø€ ðF ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐÝ+¨J¸	ÑBÔBˆÝ�U 7¨\Ñ#9Ñ:Ô:Ñ;Ô;ˆØ :Ñ-ˆà#�^ FÐ1Ð1¨bÐ1Ð1ˆŒ
Ý" 6¨>ÐOÐOÀqÐOÈBÐOÐOˆŒ
Ø#�^ NÐ9Ð9°bÐ9Ð9ˆŒ
Ý"ØØð
ð ØØ˜a”[Øð
ð 
ð ð
ð 
ˆŒ
ð ð 	$Ý�e F¨XÑ$5Ñ6Ô6Ñ7Ô7ˆKÝ˜~Ðb¸;ÐR[ÐbÐbÐ_aÐbÐbˆDŒGˆGå”k‘m”mˆDŒGØ#�^ NÐ9Ð9°bÐ9Ð9ˆŒ
Ý" >°7ÐPÐPÈÐPÈRÐPÐPˆŒ
Ý)ØØØð	
ð ØØØð	
ð 	
ð ð	
ð 	
ˆŒð 6DÀaÒ5GÐ5G� .Ñ1Ô1Ð1ÍRÌ[É]Ì]ˆŒˆˆrL   rP   c                 ó   — dS )z?Zero-initialize the last batch norm (no-op for pre-activation).NrK   rÄ   s    rM   rÅ   zPreBottleneck.zero_init_lastÈ  s   € àˆrL   rÆ   c                 ó”  — |                       |¦  «        }|}|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        �+|                      |¦  «        |                      |¦  «        z   }|S rÈ   )	rÖ   r´   r×   rµ   r¶   rØ   r·   r:   r¹   rÉ   s      rM   rË   zPreBottleneck.forwardÌ  s£   € ð �JŠJ�q‰MŒMˆØˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØ�GŠG�A‰JŒJˆØ�JŠJ�q‰MŒMˆØ�JŠJ�q‰MŒMˆØŒ?Ð&ð —’˜qÑ!Ô! D§O¢O°HÑ$=Ô$=Ñ=ˆAØˆrL   rÌ   rÍ   rÒ   s   @rM   rÔ   rÔ   z  s{  ø€ € € € € ðð ð Ø(.Ø"#ØØ"Ø'Ø$Ø)+¬Ø*,¬.Ø48Ø$&ØØð!F[ð F[àðF[ð ðF[ð ð	F[ð
 ˜C ˜H”oðF[ð  ðF[ð ðF[ð ðF[ð ðF[ð ðF[ð ˜BœI”ðF[ð ˜RœYœðF[ð !  b¤i¤Ô1ðF[ð "ðF[ð F[ð F[ð F[ð F[ð F[ðPð ð ð ð˜œð ¨%¬,ð ð ð ð ð ð ð ð rL   rÔ   c                   ó˜   ‡ — e Zd ZdZdefdedededededeee                  d	e	e
j                 fˆ fd
„Zdej        dej        fd„Zˆ xZS )ÚRegStagez—RegNet stage (sequence of blocks with the same output shape).

    A stage consists of multiple bottleneck blocks with the same output dimensions.
    Nr)   rˆ   r‰   r‹   rŒ   Údrop_path_ratesÚblock_fnc                 óJ  •— t          ¦   «                              ¦   «          d| _        |dv rdnd}	t          |¦  «        D ]b}
|
dk    r|nd}|
dk    r|n|}|	|f}|�||
         nd}d                     |
dz   ¦  «        }|                      | |||f|||d	œ|¤Ž¦  «         |}	ŒcdS )
a˜  Initialize RegNet stage.

        Args:
            depth: Number of blocks in stage.
            in_chs: Input channels.
            out_chs: Output channels.
            stride: Stride for first block.
            dilation: Dilation rate.
            drop_path_rates: Drop path rates for each block.
            block_fn: Block class to use.
            **block_kwargs: Additional block arguments.
        F)r   r”   r   r”   r   Nr4   zb{})r‹   rŒ   r«   )r²   r³   Úgrad_checkpointingr€   ÚformatÚ
add_module)rº   r)   rˆ   r‰   r‹   rŒ   rÞ   rß   Úblock_kwargsÚfirst_dilationÚiÚblock_strideÚblock_in_chsÚblock_dilationÚdprÚnamer¾   s                   €rM   r³   zRegStage.__init__ê  s   ø€ õ. 	‰Œ×ÒÑÔÐØ"'ˆÔà&¨&Ð0Ð0˜˜°aˆÝ�u‘”ð 	&ð 	&ˆAØ%&¨!¢V V˜6˜6°ˆLØ%&¨!¢V V˜6˜6°ˆLØ,¨hÐ7ˆNØ(7Ð(C�/ !Ô$Ð$ÈˆCØ—<’<  A¡Ñ&Ô&ˆDØ�OŠOØØ�Ø Øðð (Ø+Ø#&ðð ð #ðð ñ
ô 
ð 
ð &ˆNˆNð#	&ð 	&rL   rÆ   rP   c                 óÚ   — | j         rAt          j                             ¦   «         s#t	          |                      ¦   «         |¦  «        }n"|                      ¦   «         D ]} ||¦  «        }Œ|S )z�Forward pass through all blocks in the stage.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        )rá   ru   ÚjitÚis_scriptingr    Úchildren)rº   rÆ   Úblocks      rM   rË   zRegStage.forward  sh   € ð Ô"ð 	­5¬9×+AÒ+AÑ+CÔ+Cð 	Ý˜tŸ}š}™œ°Ñ2Ô2ˆAˆAàŸš™œð ð �Ø�E˜!‘H”H��ØˆrL   )rB   rC   rD   rE   r¨   rF   r
   r	   rH   r   r�   rÏ   r³   ru   rÐ   rË   rÑ   rÒ   s   @rM   rÝ   rÝ   ä  sÎ   ø€ € € € € ðð ð 6:Ø(2ð,&ð ,&àð,&ð ð,&ð ð	,&ð
 ð,&ð ð,&ð & d¨5¤kÔ2ð,&ð ˜2œ9”oð,&ð ,&ð ,&ð ,&ð ,&ð ,&ð\˜œð ¨%¬,ð ð ð ð ð ð ð ð rL   rÝ   c                   ó  ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 	 d-d	ed
ededededededefˆ fd„Z		 	 	 d.d	edededede
eeeef                  eeef         f         f
d„Zej        j        d/dedeeef         fd„¦   «         Zej        j        d0deddfd„¦   «         Zej        j        dej        fd„¦   «         Zd1dedee         ddfd„Z	 	 	 	 	 d2dej        deeeee         f                  d ed!ed"ed#edeeej                 e
ej        eej                 f         f         fd$„Z	 	 	 d3deeee         f         d&ed'edee         fd(„Zdej        dej        fd)„Zd/dej        d*edej        fd+„Zdej        dej        fd,„Zˆ xZ S )4r%   z¬RegNet-X, Y, and Z Models.

    Paper: https://arxiv.org/abs/2003.13678
    Original Impl: https://github.com/facebookresearch/pycls/blob/master/pycls/models/regnet.py
    r®   éè  r7   r¢   r4   TNÚcfgÚin_chansÚnum_classesÚoutput_strideÚglobal_poolÚ	drop_rater«   rÅ   c           
      ó°  •— t          ¦   «                              ¦   «          |	|
dœ}|| _        || _        || _        |dv sJ ‚t          |fi |¤Ž}|j        }t          |j        |j	        ¬¦  «        }|j
        rt          ||dfddi|¤Ž| _        nt          ||dfddi|¤|¤Ž| _        t          |dd¬¦  «        g| _        |}d}|                      |||¬	¦  «        \  }}t!          |¦  «        d
k    sJ ‚|j
        rt"          nt$          }t'          |¦  «        D ]x\  }}d                     |dz   ¦  «        }|                      |t-          d||dœ|¤|¤|¤Ž¦  «         |d         }||d         z  }| xj        t          |||¬¦  «        gz  c_        Œy|j        r)t          ||j        fddi|¤|¤Ž| _        |j        | _        nK|j        p|j
        }|r t5          |j        ¦  «        ¦   «         nt7          j        ¦   «         | _        || _        | j        | _        t=          d| j        |||dœ|¤Ž| _        tA          tC          tD          |¬¦  «        | ¦  «         dS )a  Initialize RegNet model.

        Args:
            cfg: Model architecture configuration.
            in_chans: Number of input channels.
            num_classes: Number of classifier classes.
            output_stride: Output stride of network, one of (8, 16, 32).
            global_pool: Global pooling type.
            drop_rate: Dropout rate.
            drop_path_rate: Stochastic depth drop-path rate.
            zero_init_last: Zero-init last weight of residual path.
            kwargs: Extra kwargs overlayed onto cfg.
        rp   )ri   é   r7   r­   r®   r‹   r”   Ústem)Únum_chsÚ	reductionÚmodule)rö   r«   é   zs{}r   )rˆ   rß   r‰   rŠ   )Úin_featuresrõ   Ú	pool_typerø   )rÅ   NrK   )#r²   r³   rõ   rô   rø   r   r8   r¤   r?   rA   r<   r   rû   r   Úfeature_infoÚ_get_stage_argsr{   rÔ   r¨   Ú	enumeraterâ   rã   rÝ   r=   Ú
final_convr;   r   r�   r—   Úhead_hidden_sizer   Úheadr!   r   Ú_init_weights)rº   ró   rô   rõ   rö   r÷   rø   r«   rÅ   rq   rr   Úkwargsr‘   r8   Úna_argsÚ
prev_widthÚcurr_strideÚper_stage_argsÚcommon_argsrß   ræ   Ú
stage_argsÚ
stage_nameÚ	final_actr¾   s                           €rM   r³   zRegNet.__init__0  s  ø€ õ6 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ&ˆÔØ ˆŒØ"ˆŒØ Ð+Ð+Ð+Ð+Ý�cÐ$Ð$˜VÐ$Ð$ˆð ”^ˆ
Ý ¤¸3¼>ÐJÑJÔJˆØŒ:ð 	XÝ% h°
¸AÐNÐNÀaÐNÈ2ÐNÐNˆDŒIˆIå# H¨j¸!ÐWÐWÀAÐWÈÐWÐTVÐWÐWˆDŒIÝ!¨*ÀÈ&ÐQÑQÔQÐRˆÔð  ˆ
ØˆØ&*×&:Ò&:ØØ'Ø)ð ';ñ '
ô '
Ñ#ˆ˜õ
 �>Ñ"Ô" aÒ'Ð'Ð'Ð'Ø$'¤JÐ>•=�=µJˆÝ& ~Ñ6Ô6ð 	fð 	f‰MˆAˆzØŸš a¨!¡eÑ,Ô,ˆJØ�OŠOØÝð Ø%Ø%ðð ð !ðð "ð	ð
 ðð ñ	ô 	ð 	ð $ IÔ.ˆJØ˜: hÔ/Ñ/ˆKØÐÔ¥$¨zÀ[ÐYcÐ"dÑ"dÔ"dÐ!eÑeÐÔÐð Ôð 	+Ý)¨*°cÔ6FÐgÐgÐTUÐgÐY`ÐgÐdfÐgÐgˆDŒOØ #Ô 0ˆDÔÐàœÐ4¨#¬*ˆIØ@IÐ\Ð:�m¨C¬MÑ:Ô:Ñ<Ô<Ð<ÍrÌ{É}Ì}ˆDŒOØ *ˆDÔØ $Ô 1ˆÔÝ"ð 
ØÔ)Ø#Ø!Øð	
ð 
ð
 ð
ð 
ˆŒ	õ 	•G�M¸.ÐIÑIÔIÈ4ÑPÔPÐPÐPÐPrL   r”   Údefault_striderP   c           
      óþ  ‡‡— t          ‰j        ‰j        ‰j        ‰j        ‰j        ¦  «        \  }}}t          j        t          j        |d¬¦  «        d¬¦  «        \  }}	| 	                    ¦   «         |	 	                    ¦   «         }	}ˆfd„t          |¦  «        D ¦   «         }
g }g }d}d}t          |¦  «        D ]A}||k    r||z  }d}n|}||z  }|                     |¦  «         |                     |¦  «         ŒBt          ||	d¬¦  «        }t          ||
|‰j        ¬	¦  «        \  }}g d
¢Šˆfd„t          ||||	|
||¦  «        D ¦   «         }t!          ‰j        ‰j        ‰j        ‰j        ‰j        ¬¦  «        }||fS )aM  Generate stage arguments from configuration.

        Args:`
            cfg: RegNet configuration.
            default_stride: Default stride for stages.
            output_stride: Target output stride.
            drop_path_rate: Stochastic depth rate.

        Returns:
            Tuple of (per_stage_args, common_args).
        ro   )rq   T)Úreturn_countsc                 ó   •— g | ]	}‰j         ‘Œ
S rK   )r3   )r[   rt   ró   s     €rM   r^   z*RegNet._get_stage_args.<locals>.<listcomp>¡  s   ø€ Ð@Ð@Ð@¨�CÔ$Ð@Ð@Ð@rL   r”   r   )Ú	stagewise)rW   )r‰   r‹   rŒ   r)   r3   r1   rÞ   c                 óJ   •— g | ]}t          t          ‰|¦  «        ¦  «        ‘Œ S rK   )r¤   rf   )r[   ÚparamsÚ	arg_namess     €rM   r^   z*RegNet._get_stage_args.<locals>.<listcomp>´  s8   ø€ ð 
ð 
ð 
Ø-3�D•�Y Ñ'Ô'Ñ(Ô(ð
ð 
ð 
rL   )r:   r5   r;   r?   rA   )r‡   r-   r+   r/   r)   r1   ru   r|   r   r‚   r€   Úappendr   rh   r6   rf   r¤   r:   r5   r;   r?   rA   )rº   ró   r  rö   r«   rT   r…   Ústage_gsÚstage_widthsÚstage_depthsÚstage_brÚstage_stridesÚstage_dilationsÚ
net_striderŒ   rt   r‹   Ú	stage_dprr  r  r  s    `                  @rM   r  zRegNet._get_stage_args‰  sÞ  øø€ õ& (7°s´v¸s¼vÀsÄvÈsÌyÐZ]ÔZhÑ'iÔ'iÑ$ˆ�
˜Hõ &+¤\µ%´,¸vÈeÐ2TÑ2TÔ2TÐdhÐ%iÑ%iÔ%iÑ"ˆ�lØ%1×%8Ò%8Ñ%:Ô%:¸L×<OÒ<OÑ<QÔ<Q�lˆØ@Ð@Ð@Ð@­e°JÑ.?Ô.?Ð@Ñ@Ô@ˆØˆØˆØˆ
ØˆÝ�zÑ"Ô"ð 	-ð 	-ˆAØ˜]Ò*Ð*Ø˜NÑ*�Ø��à'�Ø˜fÑ$�
Ø× Ò  Ñ(Ô(Ð(Ø×"Ò" 8Ñ,Ô,Ð,Ð,Ý-¨n¸lÐVZÐ[Ñ[Ô[ˆ	å!:Ø˜( H¸Ô8Kð"Mñ "Mô "MÑˆ�hàoÐoÐoˆ	ð
ð 
ð 
ð 
å�˜m¨_¸lÈHÐV^Ð`iÑjÔjð
ñ 
ô 
ˆõ Ø”~Ø”\Ø”~Ø”mØ”~ð
ñ 
ô 
ˆð ˜{Ð*Ð*rL   FÚcoarsec                 ó,   — t          d|rdnd¬¦  «        S )z"Group parameters for optimization.z^stemz^s(\d+)z^s(\d+)\.b(\d+))rû   Úblocks)r¤   )rº   r#  s     rM   Úgroup_matcherzRegNet.group_matcherÁ  s+   € õ ØØ!'Ð?�:�:Ð-?ð
ñ 
ô 
ð 	
rL   Úenablec                 ól   — t          |                      ¦   «         ¦  «        dd…         D ]	}||_        Œ
dS )z)Enable or disable gradient checkpointing.r   éÿÿÿÿN)Úlistrï   rá   )rº   r'  Úss      rM   Úset_grad_checkpointingzRegNet.set_grad_checkpointingÉ  sA   € õ �d—m’m‘o”oÑ&Ô& q¨ tÔ,ð 	*ð 	*ˆAØ#)ˆAÔ Ð ð	*ð 	*rL   c                 ó   — | j         j        S )zGet the classifier head.)r  ÚfcrÄ   s    rM   Úget_classifierzRegNet.get_classifierÏ  s   € ð ŒyŒ|ÐrL   c                 ó\   — t          | ¦  «        }|| _         | j        j        |fd|i|¤Ž dS )zŸReset the classifier head.

        Args:
            num_classes: Number of classes for new classifier.
            global_pool: Global pooling type.
        r  N)r   rõ   r  Úreset)rº   rõ   r÷   r‘   s       rM   Úreset_classifierzRegNet.reset_classifierÔ  s@   € õ ˜dÑ#Ô#ˆØ&ˆÔØˆŒ	Œ˜ÐAÐA¨{ÐA¸bÐAÐAÐAÐAÐArL   ÚNCHWrÆ   ÚindicesÚnormÚ
stop_earlyÚ
output_fmtÚintermediates_onlyc                 ó‚  — |dv s
J d¦   «         ‚g }t          d|¦  «        \  }}	d}
|                      |¦  «        }|
|v r|                     |¦  «         d}|r
|d|	…         }|D ]9}|
dz  }
 t          | |¦  «        |¦  «        }|
|v r|                     |¦  «         Œ:|r|S |
dk    r|                      |¦  «        }||fS )	aï   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:

        )r3  zOutput shape must be NCHW.é   r   ©Ús1Ús2Ús3Ús4Nr   rÿ   )r   rû   r  Úgetattrr  )rº   rÆ   r4  r5  r6  r7  r8  ÚintermediatesÚtake_indicesÚ	max_indexÚfeat_idxÚlayer_namesÚns                rM   Úforward_intermediateszRegNet.forward_intermediatesß  s  € ð* ˜YÐ&Ð&Ð&Ð(DÑ&Ô&Ð&ØˆÝ"6°q¸'Ñ"BÔ"BÑˆ�ið ˆØ�IŠI�a‰LŒLˆØ�|Ð#Ð#Ø× Ò  Ñ#Ô#Ð#à.ˆØð 	2Ø% j y jÔ1ˆKØð 	(ð 	(ˆAØ˜‰MˆHØ •˜˜aÑ Ô  Ñ#Ô#ˆAØ˜<Ð'Ð'Ø×$Ò$ QÑ'Ô'Ð'øàð 	!Ø Ð à�qŠ=ˆ=Ø—’ Ñ"Ô"ˆAà�-ÐÐrL   r   Ú
prune_normÚ
prune_headc                 óþ   — t          d|¦  «        \  }}d}||d…         }|D ]$}t          | |t          j        ¦   «         ¦  «         Œ%|dk     rt          j        ¦   «         | _        |r|                      dd¦  «         |S )aE  Prune layers not required for specified intermediates.

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

        Returns:
            List of indices that were kept.
        r:  r;  Nrÿ   r   r£   )r   Úsetattrr�   r—   r  r2  )rº   r4  rH  rI  rB  rC  rE  rF  s           rM   Úprune_intermediate_layersz RegNet.prune_intermediate_layers  s�   € õ  #7°q¸'Ñ"BÔ"BÑˆ�iØ.ˆØ! ) * *Ô-ˆØð 	,ð 	,ˆAÝ�D˜!�Rœ[™]œ]Ñ+Ô+Ð+Ð+Ø�qŠ=ˆ=Ý œk™mœmˆDŒOØð 	)Ø×!Ò! ! RÑ(Ô(Ð(ØÐrL   c                 ó  — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|S )z’Forward pass through feature extraction layers.

        Args:
            x: Input tensor.

        Returns:
            Feature tensor.
        )rû   r<  r=  r>  r?  r  ©rº   rÆ   s     rM   Úforward_featureszRegNet.forward_features*  se   € ð �IŠI�a‰LŒLˆØ�GŠG�A‰JŒJˆØ�GŠG�A‰JŒJˆØ�GŠG�A‰JŒJˆØ�GŠG�A‰JŒJˆØ�OŠO˜AÑÔˆØˆrL   Ú
pre_logitsc                 ó^   — |r|                       ||¬¦  «        n|                       |¦  «        S )zàForward pass through classifier head.

        Args:
            x: Input features.
            pre_logits: Return features before final linear layer.

        Returns:
            Classification logits or features.
        )rP  )r  )rº   rÆ   rP  s      rM   Úforward_headzRegNet.forward_head;  s0   € ð 7AÐRˆt�yŠy˜ zˆyÑ2Ô2Ð2ÀdÇiÂiÐPQÁlÄlÐRrL   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S )zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output logits.
        )rO  rR  rN  s     rM   rË   zRegNet.forwardG  s/   € ð ×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆrL   )	r®   rò   r7   r¢   r4   r4   TNN)r”   r7   r4   ©F)T)N)NFFr3  F)r   FT)!rB   rC   rD   rE   r&   rF   rI   rH   rJ   r³   r   r	   r   r   r  ru   rí   Úignorer&  r,  r�   rÏ   r/  r
   r2  rÐ   r   rG  rL  rO  rR  rË   rÑ   rÒ   s   @rM   r%   r%   )  s¶  ø€ € € € € ðð ð Ø#Ø!#Ø$Ø!Ø$&Ø#'ØØðWQð WQàðWQð ðWQð ð	WQð
 ðWQð ðWQð ðWQð "ðWQð !ðWQð WQð WQð WQð WQð WQðx #$Ø!#Ø$&ð6+ð 6+àð6+ð  ð6+ð ð	6+ð
 "ð6+ð 
ˆt�D˜˜c˜”NÔ# T¨#¨s¨(¤^Ð3Ô	4ð6+ð 6+ð 6+ð 6+ðp „YÔð
ð 
 Dð 
°T¸#¸s¸(´^ð 
ð 
ð 
ñ Ôð
ð „YÔð*ð *¨Tð *¸Tð *ð *ð *ñ Ôð*ð
 „YÔð ¤	ð ð ð ñ Ôðð	Bð 	B¨Cð 	B¸hÀs¼mð 	BÐW[ð 	Bð 	Bð 	Bð 	Bð 8<ØØ$Ø$Ø',ð. ð . àŒ|ð. ð ˜e C¨¨c¬ NÔ3Ô4ð. ð ð	. ð
 ð. ð ð. ð !%ð. ð 
ˆt�E”LÔ! 5¨¬°t¸E¼LÔ7IÐ)IÔ#JÐJÔ	Kð. ð . ð . ð . ðd ./Ø$Ø#ð	ð à˜3  S¤	˜>Ô*ðð ðð ð	ð
 
ˆcŒðð ð ð ð6 %¤,ð °5´<ð ð ð ð ð"
Sð 
S˜eœlð 
S¸ð 
SÈÌð 
Sð 
Sð 
Sð 
Sð˜œð ¨%¬,ð ð ð ð ð ð ð ð rL   r£   rþ   rë   rÅ   c                 ó„  — t          | t          j        ¦  «        r‹| j        d         | j        d         z  | j        z  }|| j        z  }| j        j                             dt          j
        d|z  ¦  «        ¦  «         | j        � | j        j                             ¦   «          dS dS t          | t          j        ¦  «        rVt          j                             | j        dd¬¦  «         | j        �&t          j                             | j        ¦  «         dS dS |r&t!          | d¦  «        r|                      ¦   «          dS dS dS )	z¯Initialize module weights.

    Args:
        module: PyTorch module to initialize.
        name: Module name.
        zero_init_last: Zero-initialize last layer weights.
    r   r   ç       @Nr4   g{®Gáz„?)ÚmeanÚstdrÅ   )Ú
isinstancer�   ÚConv2drŠ   Úout_channelsrV   rÃ   ÚdataÚnormal_ry   ÚsqrtÚbiasÚzero_ÚLinearrÀ   rÁ   ÚhasattrrÅ   )rþ   rë   rÅ   Úfan_outs       rM   r  r  U  s>  € õ �&�"œ)Ñ$Ô$ð  ØÔ$ QÔ'¨&Ô*<¸QÔ*?Ñ?À&ÔBUÑUˆØ�F”MÑ!ˆØŒÔ×"Ò" 1¥d¤i°°g±Ñ&>Ô&>Ñ?Ô?Ð?ØŒ;Ð"ØŒKÔ×"Ò"Ñ$Ô$Ð$Ð$Ð$ð #Ð"å	�F�BœIÑ	&Ô	&ð  Ý
Œ�Š˜œ¨C°TˆÑ:Ô:Ð:ØŒ;Ð"ÝŒG�NŠN˜6œ;Ñ'Ô'Ð'Ð'Ð'ð #Ð"à	ð  �G FÐ,<Ñ=Ô=ð  Ø×ÒÑÔÐÐÐð ð  ð  ð  rL   Ú
state_dictc                 ó„  — |                       d| ¦  «        } g d¢}d| v rþddl}| d         d         d         } i }| d                              ¦   «         D ]ƒ\  }}|                     dd	¦  «        }|                     d
d¦  «        }|                     dd„ |¦  «        }|                     dd|¦  «        }|D ]\  }}|                     ||¦  «        }Œ|||<   Œ„| d                              ¦   «         D ])\  }}d|v sd|v rŒ|                     dd¦  «        }|||<   Œ*|S d| v rŸddl}i }|                      ¦   «         D ]‚\  }}|                     dd	¦  «        }|                     dd¦  «        }|                     dd„ |¦  «        }|D ]\  }}|                     ||¦  «        }Œ|                     dd¦  «        }|||<   Œƒ|S | S )zžFilter and remap state dict keys for compatibility.

    Args:
        state_dict: Raw state dictionary.

    Returns:
        Filtered state dictionary.
    Úmodel))zf.a.0z
conv1.conv)zf.a.1zconv1.bn)zf.b.0z
conv2.conv)zf.b.1zconv2.bn)z
f.final_bnúconv3.bn)zf.se.excitation.0zse.fc1)zf.se.excitation.2zse.fc2)zf.ser¶   )zf.c.0ú
conv3.conv)zf.c.1rh  )zf.cri  )zproj.0údownsample.conv)zproj.1zdownsample.bn)Úprojrj  Úclassy_state_dictr   NÚ
base_modelÚtrunkz_feature_blocks.conv1.stem.0ú	stem.convz_feature_blocks.conv1.stem.1zstem.bnz&^_feature_blocks.res\d.block(\d)-(\d+)c                 ó˜   — dt          |                      d¦  «        ¦  «        › dt          |                      d¦  «        ¦  «        dz   › �S )Nr+  r   ú.br”   ©rF   Úgroup©rÆ   s    rM   ú<lambda>z_filter_fn.<locals>.<lambda>�  ó<   € ÐF�c !§'¢'¨!¡*¤*™oœoÐFÐFµ°Q·W²W¸Q±Z´Z±´À1Ñ1DÐFÐF€ rL   zs(\d)\.b(\d+)\.bnzs\1.b\2.downsample.bnÚheadsÚprojection_headÚ
prototypesz0.clf.0úhead.fczstem.0.weightzstem.0zstem.1z)trunk_output.block(\d)\.block(\d+)\-(\d+)c                 ó˜   — dt          |                      d¦  «        ¦  «        › dt          |                      d¦  «        ¦  «        dz   › �S )Nr+  r   rq  r®   rr  rt  s    rM   ru  z_filter_fn.<locals>.<lambda>£  rv  rL   zfc.zhead.fc.)ÚgetÚreÚitemsr   Úsub)re  Úreplacesr}  ÚoutÚkÚvr+  Úrs           rM   Ú
_filter_fnr…  k  s<  € ð —’ ¨Ñ4Ô4€Jðð ð €Hð  ˜jÐ(Ð(àˆ	ˆ	ˆ	ØÐ 3Ô4°\ÔBÀ7ÔKˆ
ØˆØ˜wÔ'×-Ò-Ñ/Ô/ð 		ð 		‰DˆAˆqØ—	’	Ð8¸+ÑFÔFˆAØ—	’	Ð8¸)ÑDÔDˆAØ—’Ø9ØFÐFÈñKô KˆAð —’Ð+Ð-EÀqÑIÔIˆAØ ð $ð $‘��1Ø—I’I˜a ‘O”O��ØˆC�‰FˆFØ˜wÔ'×-Ò-Ñ/Ô/ð 	ð 	‰DˆAˆqØ  AÐ%Ð%¨¸Ð):Ð):ØØ—	’	˜) YÑ/Ô/ˆAØˆC�‰FˆFØˆ
Ø˜*Ð$Ð$àˆ	ˆ	ˆ	ØˆØ×$Ò$Ñ&Ô&ð 		ð 		‰DˆAˆqØ—	’	˜( KÑ0Ô0ˆAØ—	’	˜( IÑ.Ô.ˆAØ—’Ø<ØFÐFÈñKô KˆAð !ð $ð $‘��1Ø—I’I˜a ‘O”O��Ø—	’	˜% Ñ,Ô,ˆAØˆC�‰FˆFØˆ
ØÐrL   Úregnetx_002r0   g¸…ëQ8B@gìQ¸…ë@é   )r+   r-   r/   r1   r)   Úregnetx_004g{®Gáz8@gR¸…ëQ@rú   é   Úregnetx_004_tvgÍÌÌÌÌÌì?)r+   r-   r/   r1   r)   r6   Úregnetx_006é0   g\�Âõ(|B@gìQ¸…ë@Úregnetx_008é8   g=
×£pÝA@g=
×£p=@Úregnetx_016r*   gáz®GA@g      @é   Úregnetx_032éX   g�Âõ(\O:@é   Úregnetx_040é`   g33333SC@gq=
×£p@é(   é   Úregnetx_064é¸   g
×£p=jN@g�Âõ(\� @é   Úregnetx_080gHáz®ÇH@g
×£p=
@éx   Úregnetx_120é¨   g×£p=
WR@gö(\�Âõ@ép   é   Úregnetx_160éØ   gìQ¸…ËK@gÍÌÌÌÌÌ @é€   Úregnetx_320é@  g×£p=
wQ@rW  Úregnety_002r©   )r+   r-   r/   r1   r)   r5   Úregnety_004g¤p=
×ã;@g¸…ëQ¸ @Úregnety_006g…ëQ¸E@@g�Âõ(\�@é   Úregnety_008gìQ¸…kC@g333333@é   Úregnety_008_tv)r+   r-   r/   r1   r)   r5   r6   Úregnety_016gö(\�Âµ4@g333333@é   Úregnety_032r,   r.   r(   Úregnety_040g)\�Âõh?@é@   Úregnety_064g\�Âõ(œ@@g)\�Âõ(@éH   Úregnety_080éÀ   g®Gáz4S@g…ëQ¸…@Úregnety_080_tvÚregnety_120Úregnety_160éÈ   g…ëQ¸ŽZ@g×£p=
×@Úregnety_320éè   g)\�Âõø\@g=
×£p=@é   Úregnety_640i`  g�Âõ(\ob@iH  Úregnety_1280iÈ  gÃõ(\�d@g)\�Âõ(@i  Úregnety_2560i€  gÃõ(\�Úl@iu  Úregnety_040_sgnÚsilu)r1   )r+   r-   r/   r1   r)   r5   r?   rA   Úregnetv_040T)r)   r+   r-   r/   r1   r5   r<   r?   Úregnetv_064r¢   )	r)   r+   r-   r/   r1   r5   r<   r?   r:   Úregnetz_005gffffff%@g®Gáz@rÿ   g      @i   )r)   r+   r-   r/   r1   r3   r5   r:   r;   r=   r?   Úregnetz_040é   g      -@g+‡ÙÎ@Úregnetz_040_hi   ÚvariantÚ
pretrainedc                 óP   — t          t          | |ft          |          t          dœ|¤ŽS )zÓCreate a RegNet model.

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

    Returns:
        RegNet model instance.
    )Ú	model_cfgÚpretrained_filter_fn)r   r%   Ú
model_cfgsr…  )rÈ  rÉ  r	  s      rM   Ú_create_regnetrÎ  í  s<   € õ  Ý�˜ðå˜WÔ%Ý'ðð ð ð	ð ð rL   Úurlc                 ó:   — | dddddddt           t          dd	d
dœ|¥S )z»Create default configuration dictionary.

    Args:
        url: Model weight URL.
        **kwargs: Additional configuration options.

    Returns:
        Configuration dictionary.
    rò   ©r®   éà   rÒ  ©é   rÔ  )r®   é   rÕ  gffffffî?r2   Úbicubicro  rz  z
apache-2.0)rÏ  rõ   Ú
input_sizeÚ	pool_sizeÚtest_input_sizeÚcrop_pctÚtest_crop_pctÚinterpolationrX  rY  Ú
first_convÚ
classifierÚlicenser   ©rÏ  r	  s     rM   Ú_cfgrá  ÿ  s?   € ð  4°}ÐSYØ(°dÈSØ"Õ,AÕJ^Ø!°Øðð ð
 $*ðð rL   c                 ó8   — | dddddt           t          dddd	d
œ|¥S )z¹Create pycls configuration dictionary.

    Args:
        url: Model weight URL.
        **kwargs: Additional configuration options.

    Returns:
        Configuration dictionary.
    rò   rÑ  rÓ  g      ì?rÖ  ro  rz  Úmitz)https://github.com/facebookresearch/pycls©rÏ  rõ   r×  rØ  rÚ  rÜ  rX  rY  rÝ  rÞ  rß  Ú
origin_urlr   rà  s     rM   Ú_cfgpycræ    s?   € ð  4°}ÐSYØ¨IÝ%Õ.BØ!°ØÐ(Sðð ð
 X^ðð rL   c                 ó8   — | dddddt           t          dddd	d
œ|¥S )zÂCreate torchvision v2 configuration dictionary.

    Args:
        url: Model weight URL.
        **kwargs: Additional configuration options.

    Returns:
        Configuration dictionary.
    rò   rÑ  rÓ  gáz®Gáî?rÖ  ro  rz  zbsd-3-clausez!https://github.com/pytorch/visionrä  r   rà  s     rM   Ú_cfgtv2rè  %  s?   € ð  4°}ÐSYØ¨IÝ%Õ.BØ!°Ø!Ð1Tðð ð
 Y_ðð rL   zregnety_032.ra_in1kztimm/znhttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-weights/regnety_032_ra-7f2439f9.pth)Ú	hf_hub_idrÏ  zregnety_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_040_ra3-670e1166.pthzregnety_064.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_064_ra3-aa26dc7d.pthzregnety_080.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnety_080_ra3-1fdc4344.pthzregnety_120.sw_in12k_ft_in1k)ré  zregnety_160.sw_in12k_ft_in1kzregnety_160.lion_in12k_ft_in1kzregnety_120.sw_in12ki-.  )ré  rõ   zregnety_160.sw_in12kzregnety_040_sgn.untrained)rÏ  zregnetv_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetv_040_ra3-c248f51f.pthrû   )ré  rÏ  rÝ  zregnetv_064.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetv_064_ra3-530616c2.pthzregnetz_005.untrainedzregnetz_040.ra3_in1kzshttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetz_040_ra3-9007edf5.pth)r®   é   rê  )ri   ri   r2   )r®   r¥  r¥  )ré  rÏ  r×  rØ  rÚ  rÙ  zregnetz_040_h.ra3_in1kzthttps://github.com/huggingface/pytorch-image-models/releases/download/v0.1-tpu-weights/regnetz_040h_ra3-f594343b.pthzregnety_160.deit_in1kz<https://dl.fbaipublicfiles.com/deit/regnety_160-a5fe301d.pthzregnetx_004_tv.tv2_in1kz?https://download.pytorch.org/models/regnet_x_400mf-62229a5f.pthzregnetx_008.tv2_in1kz?https://download.pytorch.org/models/regnet_x_800mf-94a99ebd.pthzregnetx_016.tv2_in1kz?https://download.pytorch.org/models/regnet_x_1_6gf-a12f2b72.pthzregnetx_032.tv2_in1kz?https://download.pytorch.org/models/regnet_x_3_2gf-7071aa85.pthzregnetx_080.tv2_in1kz=https://download.pytorch.org/models/regnet_x_8gf-2b70d774.pthzregnetx_160.tv2_in1kz>https://download.pytorch.org/models/regnet_x_16gf-ba3796d7.pthzregnetx_320.tv2_in1kz>https://download.pytorch.org/models/regnet_x_32gf-6eb8fdc6.pthzregnety_004.tv2_in1kz?https://download.pytorch.org/models/regnet_y_400mf-e6988f5f.pthzregnety_008_tv.tv2_in1kz?https://download.pytorch.org/models/regnet_y_800mf-58fc7688.pthzregnety_016.tv2_in1kz?https://download.pytorch.org/models/regnet_y_1_6gf-0d7bc02a.pthzregnety_032.tv2_in1kz?https://download.pytorch.org/models/regnet_y_3_2gf-9180c971.pthzregnety_080_tv.tv2_in1kz=https://download.pytorch.org/models/regnet_y_8gf-dc2b1b54.pthzregnety_160.tv2_in1kz>https://download.pytorch.org/models/regnet_y_16gf-3e4a00f9.pthzregnety_320.tv2_in1kz>https://download.pytorch.org/models/regnet_y_32gf-8db6d4b5.pthzregnety_160.swag_ft_in1kzChttps://download.pytorch.org/models/regnet_y_16gf_swag-43afe44d.pthzcc-by-nc-4.0)r®   é€  rë  )é   rì  )ré  rÏ  rß  r×  rØ  rÚ  zregnety_320.swag_ft_in1kzChttps://download.pytorch.org/models/regnet_y_32gf_swag-04fdfa75.pthzregnety_1280.swag_ft_in1kzDhttps://download.pytorch.org/models/regnet_y_128gf_swag-c8ce3e52.pthzregnety_160.swag_lc_in1kzFhttps://download.pytorch.org/models/regnet_y_16gf_lc_swag-f3ec0043.pth)ré  rÏ  rß  zregnety_320.swag_lc_in1kzFhttps://download.pytorch.org/models/regnet_y_32gf_lc_swag-e1583746.pthzregnety_1280.swag_lc_in1kzGhttps://download.pytorch.org/models/regnet_y_128gf_lc_swag-cbe8ce12.pthzregnety_320.seer_ft_in1kzseer-licensez)https://github.com/facebookresearch/visslzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet32_finetuned_in1k_model_final_checkpoint_phase78.torch)ré  rß  rå  rÏ  r×  rØ  rÚ  zregnety_640.seer_ft_in1kzhttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet64_finetuned_in1k_model_final_checkpoint_phase78.torchzregnety_1280.seer_ft_in1kz€https://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet128_finetuned_in1k_model_final_checkpoint_phase78.torchzregnety_2560.seer_ft_in1kz€https://dl.fbaipublicfiles.com/vissl/model_zoo/seer_finetuned/seer_regnet256_finetuned_in1k_model_final_checkpoint_phase38.torchzregnety_320.seerzihttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_regnet32d/seer_regnet32gf_model_iteration244000.torch)ré  rÏ  rõ   rß  rå  zregnety_640.seerzphttps://dl.fbaipublicfiles.com/vissl/model_zoo/seer_regnet64/seer_regnet64gf_model_final_checkpoint_phase0.torchzregnety_1280.seerz¢https://dl.fbaipublicfiles.com/vissl/model_zoo/swav_ig1b_regnet128Gf_cnstant_bs32_node16_sinkhorn10_proto16k_syncBN64_warmup8k/model_final_checkpoint_phase0.torchzregnetx_002.pycls_in1kzregnetx_004.pycls_in1kzregnetx_006.pycls_in1kzregnetx_008.pycls_in1kzregnetx_016.pycls_in1kzregnetx_032.pycls_in1kzregnetx_040.pycls_in1kzregnetx_064.pycls_in1kzregnetx_080.pycls_in1kzregnetx_120.pycls_in1kzregnetx_160.pycls_in1kzregnetx_320.pycls_in1kzregnety_002.pycls_in1kzregnety_004.pycls_in1kzregnety_006.pycls_in1kzregnety_008.pycls_in1kzregnety_016.pycls_in1kzregnety_032.pycls_in1kzregnety_040.pycls_in1kzregnety_064.pycls_in1kzregnety_080.pycls_in1kzregnety_120.pycls_in1kzregnety_160.pycls_in1kzregnety_320.pycls_in1kc                 ó   — t          d| fi |¤ŽS )zRegNetX-200MFr†  ©rÎ  ©rÉ  r	  s     rM   r†  r†  ð  ó   € õ ˜-¨Ð>Ð>°vÐ>Ð>Ð>rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-400MFrˆ  rî  rï  s     rM   rˆ  rˆ  ö  rð  rL   c                 ó   — t          d| fi |¤ŽS )z+RegNetX-400MF w/ torchvision group roundingrŠ  rî  rï  s     rM   rŠ  rŠ  ü  ó   € õ Ð*¨JÐAÐA¸&ÐAÐAÐArL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-600MFr‹  rî  rï  s     rM   r‹  r‹    rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-800MFr�  rî  rï  s     rM   r�  r�    rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-1.6GFr�  rî  rï  s     rM   r�  r�    rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-3.2GFr‘  rî  rï  s     rM   r‘  r‘    rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-4.0GFr”  rî  rï  s     rM   r”  r”    rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-6.4GFr˜  rî  rï  s     rM   r˜  r˜     rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-8.0GFr›  rî  rï  s     rM   r›  r›  &  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-12GFr�  rî  rï  s     rM   r�  r�  ,  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-16GFr¡  rî  rï  s     rM   r¡  r¡  2  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetX-32GFr¤  rî  rï  s     rM   r¤  r¤  8  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-200MFr¦  rî  rï  s     rM   r¦  r¦  >  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-400MFr§  rî  rï  s     rM   r§  r§  D  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-600MFr¨  rî  rï  s     rM   r¨  r¨  J  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-800MFrª  rî  rï  s     rM   rª  rª  P  rð  rL   c                 ó   — t          d| fi |¤ŽS )z+RegNetY-800MF w/ torchvision group roundingr¬  rî  rï  s     rM   r¬  r¬  V  ró  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-1.6GFr­  rî  rï  s     rM   r­  r­  \  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-3.2GFr¯  rî  rï  s     rM   r¯  r¯  b  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-4.0GFr°  rî  rï  s     rM   r°  r°  h  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-6.4GFr²  rî  rï  s     rM   r²  r²  n  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-8.0GFr´  rî  rï  s     rM   r´  r´  t  rð  rL   c                 ó   — t          d| fi |¤ŽS )z+RegNetY-8.0GF w/ torchvision group roundingr¶  rî  rï  s     rM   r¶  r¶  z  ró  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-12GFr·  rî  rï  s     rM   r·  r·  €  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-16GFr¸  rî  rï  s     rM   r¸  r¸  †  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-32GFrº  rî  rï  s     rM   rº  rº  Œ  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-64GFr½  rî  rï  s     rM   r½  r½  ’  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-128GFr¾  rî  rï  s     rM   r¾  r¾  ˜  ó   € õ ˜.¨*Ð?Ð?¸Ð?Ð?Ð?rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-256GFr¿  rî  rï  s     rM   r¿  r¿  ž  r  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetY-4.0GF w/ GroupNorm rÀ  rî  rï  s     rM   rÀ  rÀ  ¤  s   € õ Ð+¨ZÐBÐB¸6ÐBÐBÐBrL   c                 ó   — t          d| fi |¤ŽS )zRegNetV-4.0GF (pre-activation)rÂ  rî  rï  s     rM   rÂ  rÂ  ª  rð  rL   c                 ó   — t          d| fi |¤ŽS )zRegNetV-6.4GF (pre-activation)rÃ  rî  rï  s     rM   rÃ  rÃ  °  rð  rL   c                 ó"   — t          d| fddi|¤ŽS )zÛRegNetZ-500MF
    NOTE: config found in https://github.com/facebookresearch/ClassyVision/blob/main/classy_vision/models/regnet.py
    but it's not clear it is equivalent to paper model as not detailed in the paper.
    rÄ  rÅ   Frî  rï  s     rM   rÄ  rÄ  ¶  ó!   € õ ˜-¨ÐTÐTÀEÐTÈVÐTÐTÐTrL   c                 ó"   — t          d| fddi|¤ŽS )úÛRegNetZ-4.0GF
    NOTE: config found in https://github.com/facebookresearch/ClassyVision/blob/main/classy_vision/models/regnet.py
    but it's not clear it is equivalent to paper model as not detailed in the paper.
    rÅ  rÅ   Frî  rï  s     rM   rÅ  rÅ  ¿  r  rL   c                 ó"   — t          d| fddi|¤ŽS )r  rÇ  rÅ   Frî  rï  s     rM   rÇ  rÇ  È  s!   € õ ˜/¨:ÐVÐVÀeÐVÈvÐVÐVÐVrL   Úregnetz_040h)r4   )ri   )r   r   r   NFNN)rŸ   NFNN)r£   FrK   )r£   rT  )nrE   ry   Údataclassesr   r   Ú	functoolsr   Útypingr   r   r   r	   r
   r   r   r   ru   Útorch.nnr�   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   r   r   r   r   r   Ú_builderr   Ú	_featuresr   Ú_manipulater    r!   Ú	_registryr"   r#   r$   Ú__all__r&   rH   rF   rS   rh   r‡   rÏ   rJ   r’   r™   rž   rI   r¦   r¨   rÔ   rÝ   r%   r  r…  r¤   rÍ  rÎ  rá  ræ  rè  Údefault_cfgsr†  rˆ  rŠ  r‹  r�  r�  r‘  r”  r˜  r›  r�  r¡  r¤  r¦  r§  r¨  rª  r¬  r­  r¯  r°  r²  r´  r¶  r·  r¸  rº  r½  r¾  r¿  rÀ  rÂ  rÃ  rÄ  rÅ  rÇ  rB   rK   rL   rM   ú<module>r%     sö  ððð ð2 €€€Ø *Ð *Ð *Ð *Ð *Ð *Ð *Ð *Ø Ð Ð Ð Ð Ð Ø JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ JÐ Jà €€€Ø Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ Að Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  Rð  RØ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XØ *Ð *Ð *Ð *Ð *Ð *Ø +Ð +Ð +Ð +Ð +Ð +Ø 4Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4Ð 4Ø YÐ YÐ YÐ YÐ YÐ YÐ YÐ YÐ YÐ Yà�[Ð
!€ð ð3ð 3ð 3ð 3ð 3ñ 3ô 3ñ „ð3ð&
!�eð 
! ð 
!¨ð 
!ð 
!ð 
!ð 
!ð" ñ	ð Ø�S”	ðà˜E”{ðð �S”	ðð ð	ð
 ˆ4�Œ9�d˜3”iÐÔ ðð ð ð ðD ñ>ð >Øð>àð>ð ð>ð ð	>ð
 ð>ð ð>ð ˆ4�Œ9�c˜4 œ9Ð$Ô%ð>ð >ð >ð >ðH ØØØ04ØØØñ0
ð 0
Øð0
àð0
ð ð0
ð ð	0
ð
 ð0
ð ˜T "¤)œ_Ô-ð0
ð ð0
ð „Yð0
ð 0
ð 0
ð 0
ðl ØØØ04ØØØñ&(ð &(Øð&(àð&(ð ð&(ð ð	&(ð
 ð&(ð ˜T "¤)œ_Ô-ð&(ð ð&(ð „]ð&(ð &(ð &(ð &(ð^ %+Ø04ØØØñ&ð &Ø! #œð&àð&ð ð&ð ð	&ð
 ð&ð ˜˜S˜”/ð&ð ˜T "¤)œ_Ô-ð&ð ð&ð ˆbŒiÔð&ð &ð &ð &ðRgð gð gð gð g�”ñ gô gð gðTgð gð gð gð g�B”Iñ gô gð gðTBð Bð Bð Bð BˆrŒyñ Bô Bð BðJið ið ið ið iˆRŒYñ iô ið iñX	 ð  ˜"œ)ð  ¨3ð  ÀTð  ÐVZð  ð  ð  ð  ð,>˜4  S œ>ð >¨d°3¸°8¬nð >ð >ð >ð >ðD ˆTñ =ð =ð =à�	˜R E¨d¸qÈÐKÑKÔKÐKð=ð �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �9  u°À"ÈBÐ`cÐdÑdÔdÐdð	=ð
 �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �	˜R E¨d¸rÈÐLÑLÔLÐLð=ð �	˜S U¨tÀÈ"ÐMÑMÔMÐMð=ð �	˜R E¨d¸sÈ"ÐMÑMÔMÐMð=ð �	˜S U¨tÀÈ2ÐNÑNÔNÐNð=ð �	˜S U¨s¸sÈ"ÐMÑMÔMÐMð=ð �	˜S U¨s¸sÈ"ÐMÑMÔMÐMð=ð" �	˜R E¨d¸qÈÐUYÐZÑZÔZÐZð#=ð$ �	˜R E¨d¸qÈÐUYÐZÑZÔZÐZð%=ð& �	˜R E¨d¸rÈÐVZÐ[Ñ[Ô[Ð[ð'=ð( �	˜R E¨c¸bÈÐUYÐZÑZÔZÐZð)=ð* �9  u°ÀÈ2ÐX\ÐnqÐrÑrÔrÐrð+=ð, �	˜R E¨d¸rÈÐVZÐ[Ñ[Ô[Ð[ð-=ð. �	˜R E¨d¸rÈÐVZÐ[Ñ[Ô[Ð[ð/=ð0 �	˜R E¨d¸rÈÐVZÐ[Ñ[Ô[Ð[ð1=ð2 �	˜S U¨tÀÈ"ÐW[Ð\Ñ\Ô\Ð\ð3=ð4 �	˜S U¨tÀÈ"ÐW[Ð\Ñ\Ô\Ð\ð5=ð6 �9 ¨°$À2ÈRÐZ^ÐpsÐtÑtÔtÐtð7=ð8 �	˜S U¨tÀÈ2ÐX\Ð]Ñ]Ô]Ð]ð9=ð: �	˜S V°ÀÈBÐY]Ð^Ñ^Ô^Ð^ð;=ð< �	˜S V°ÀÈBÐY]Ð^Ñ^Ô^Ð^ð==ð> �	˜S V°ÀÈ2ÐX\Ð]Ñ]Ô]Ð]ð?=ð@ �˜c f°À#ÈRÐZ^Ð_Ñ_Ô_Ð_ðA=ðB �˜c f°À#ÈRÐZ^Ð_Ñ_Ô_Ð_ðC=ðJ �IØ�%˜D¨R°rÀDØ W W¨\ÀbÐ%IÑ%IÔ%IðKñ Kô Kð KðK=ðT �	Ø�R˜E d°rÀDÐQUÐagðiñ iô ið iðU=ðX �	Ø�S˜U t¸ÀTÐRVÐbhØðñ ô ð ðY=ðb �	Ø�R˜D T°aÀcÐTXØ D°tÀvðñ ô ð ðc=ðj �	Ø�R˜D U°qÀsÐUYØ D°qÀFðñ ô ð ðk=ðr �)Ø�R˜D U°qÀsÐUYØ D°tÀvðñ ô ð ðs=€
ð@˜Cð ¨Tð Àð ð ð ð ñ$ð ˆcð  T¨#¨s¨(¤^ð ð ð ð ñ&ð �ð ¨¨S°#¨X¬ð ð ð ð ñ&ð �ð ¨¨S°#¨X¬ð ð ð ð ð& %Ð$ð u&à˜4˜4ØØ|ð~ñ ~ô ~ðu&ð
 ˜D˜DØð BðCñ Cô Cðu&ð ˜D˜DØð BðCñ Cô Cðu&ð ˜D˜DØð BðCñ Cô Cðu&ð # D D°7Ð$;Ñ$;Ô$;ðu&ð # D D°7Ð$;Ñ$;Ô$;ðu&ð  % d d°WÐ&=Ñ&=Ô&=ð!u&ð& ˜D˜DØØðñ ô ð'u&ð, ˜D˜DØØðñ ô ð-u&ð6    ¨" ¡¤ð7u&ð8 ˜D˜DØð BØðñ ô ð9u&ð@ ˜D˜DØð BØðñ ô ðAu&ðJ ˜T˜T b˜\™\œ\ðKu&ðL ˜D˜DØð BØ ¨F¸SÐR_ðañ aô aðMu&ðT ˜d˜dØð CØ ¨F¸SÐR_ðañ aô aðUu&ð` ˜T˜TØÐ]ð_ñ _ô _ðau&ðf ˜w˜wØØMð Oñ  Oô  Oðgu&ð u&ðl ˜G˜GØØMðOñ Oô Oðmu&ðr ˜G˜GØØMðOñ Oô Oðsu&ñx ˜G˜GØÙMðOñ Oô Oðyu&ñ~ ˜G˜GØÙKðMñ Mô Mðu&ñD ˜G˜GØÙLðNñ Nô NðEu&ñJ ˜G˜GØÙLðNñ Nô NðKu&ñR ˜G˜GØÙMðOñ Oô OðSu&ñX ˜w˜wØÙMð Oñ  Oô  OðYu&ñ^ ˜G˜GØÙMðOñ Oô Oð_u&ñd ˜G˜GØÙMðOñ Oô Oðeu&ñj ˜w˜wØÙKð Mñ  Mô  Mðku&ñp ˜G˜GØÙLðNñ Nô Nðqu&ñv ˜G˜GØÙLðNñ Nô Nðwu&ñ~   ØÙQÑ[iÙ ©H¸sñ!Dñ !Dô !Dðu&ñF   ØÙQÑ[iÙ ©H¸sñ!Dñ !Dô !DðGu&ñN    ØÙRÑ\jÙ ©H¸sñ"Dñ "Dô "DðOu&ñX   ØÙTÑ^lñ!nñ !nô !nðYu&ð u&ð u&ñ^   ØÙTÑ^lñ!nñ !nô !nð_u&ñd    ØÙUÑ_mñ"oñ "oô "oðeu&ñl   ØÙÑ+Vñ NÙ ©H¸sñ	!Dñ !Dô !Dðmu&ñv   ØÙÑ+Vñ NÙ ©H¸sñ	!Dñ !Dô !Dðwu&ñ@    ØÙÑ+Vñ OÙ ©H¸sñ	"Dñ "Dô "DðAu&ñJ    ØÙÑ+Vñ OÙ ©H¸sñ	"Dñ "Dô "DðKu&ñV ˜˜ØÙwØ™~Ñ:eñgñ gô gðWu&ñ^ ˜˜ØÙ~Ø™~Ñ:eñgñ gô gð_u&ñf ˜˜Øñ qØ™~Ñ:eñgñ gô gðgu&ñx ˜g˜g°Ð8Ñ8Ô8ðyu&ñz ˜g˜g°Ð8Ñ8Ô8ð{u&ñ| ˜g˜g°Ð8Ñ8Ô8ð}u&ñ~ ˜g˜g°Ð8Ñ8Ô8ðu&ñ@ ˜g˜g°Ð8Ñ8Ô8ðAu&ñB ˜g˜g°Ð8Ñ8Ô8ðCu&ñD ˜g˜g°Ð8Ñ8Ô8ðEu&ñF ˜g˜g°Ð8Ñ8Ô8ðGu&ð u&ð u&ñH ˜g˜g°Ð8Ñ8Ô8ðIu&ñJ ˜g˜g°Ð8Ñ8Ô8ðKu&ñL ˜g˜g°Ð8Ñ8Ô8ðMu&ñN ˜g˜g°Ð8Ñ8Ô8ðOu&ñR ˜g˜g°Ð8Ñ8Ô8ðSu&ñT ˜g˜g°Ð8Ñ8Ô8ðUu&ñV ˜g˜g°Ð8Ñ8Ô8ðWu&ñX ˜g˜g°Ð8Ñ8Ô8ðYu&ñZ ˜g˜g°Ð8Ñ8Ô8ð[u&ñ\ ˜g˜g°Ð8Ñ8Ô8ð]u&ñ^ ˜g˜g°Ð8Ñ8Ô8ð_u&ñ` ˜g˜g°Ð8Ñ8Ô8ðau&ñb ˜g˜g°Ð8Ñ8Ô8ðcu&ñd ˜g˜g°Ð8Ñ8Ô8ðeu&ñf ˜g˜g°Ð8Ñ8Ô8ðgu&ñh ˜g˜g°Ð8Ñ8Ô8ðiu&ð u&ñ uô u€ðp ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñBð B˜tð B¸&ð Bñ Bð Bñ „ðBð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñBð B˜tð B¸&ð Bñ Bð Bñ „ðBð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñBð B˜tð B¸&ð Bñ Bð Bñ „ðBð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ@ð @˜Tð @¸ð @ñ @ð @ñ „ð@ð
 ñ@ð @˜Tð @¸ð @ñ @ð @ñ „ð@ð
 ñCð C ð C¸6ð Cñ Cð Cñ „ðCð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñ?ð ?˜Dð ?°vð ?ñ ?ð ?ñ „ð?ð
 ñUð U˜Dð U°vð Uñ Uð Uñ „ðUð ñUð U˜Dð U°vð Uñ Uð Uñ „ðUð ñWð W˜dð W¸ð Wñ Wð Wñ „ðWð Ð ˜HÙ�Oð'ñ ô ð ð ð rL   