§
    ·¯¸j¤P  ã                   óú  — d Z ddlm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ZmZ ddlmZmZmZmZmZmZmZ dd	lmZ dd
lmZ ddlm Z  ddl!m"Z"m#Z# ddl$m%Z%m&Z& dgZ' eeej(        ¬¦  «        Z) G d„ dej*        ¦  «        Z+	 	 	 	 	 	 dCde,de,de-de-de,de-dee
e-e,e-e,f                  fd„Z.	 	 	 	 	 	 	 dDd!ee
e-e,e-e,f                  d"e-de,de-d#e-d$e/d%e/d&e,de
eej*                 eee/ef                  f         fd'„Z0 G d(„ dej*        ¦  «        Z1d)e/d*e2de1fd+„Z3dEd-e/dee/ef         fd.„Z4 e% e4d/¬0¦  «         e4d/¬0¦  «         e4d/¬0¦  «         e4d/¬0¦  «         e4d/¬0¦  «         e4¦   «          e4¦   «          e4¦   «          e4d/d1dd2d3¬4¦  «         e4d/d1dd2d3¬4¦  «         e4d/d5d1dd2d3¬6¦  «         e4d/d5d1dd2d3¬6¦  «        d7œ¦  «        Z5e&dFd*e2de1fd9„¦   «         Z6e&dFd*e2de1fd:„¦   «         Z7e&dFd*e2de1fd;„¦   «         Z8e&dFd*e2de1fd<„¦   «         Z9e&dFd*e2de1fd=„¦   «         Z:e&dFd*e2de1fd>„¦   «         Z;e&dFd*e2de1fd?„¦   «         Z<e&dFd*e2de1fd@„¦   «         Z=e&dFd*e2de1fdA„¦   «         Z>e&dFd*e2de1fdB„¦   «         Z?dS )Ga…   ReXNet

A PyTorch impl of `ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network` -
https://arxiv.org/abs/2007.00992

Adapted from original impl at https://github.com/clovaai/rexnet
Copyright (c) 2020-present NAVER Corp. MIT license

Changes for timm, feature extraction, and rounded channel variant hacked together by Ross Wightman
Copyright 2020 Ross Wightman
é    )Úpartial©Úceil)ÚAnyÚDictÚListÚOptionalÚTupleÚTypeÚUnionN©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚClassifierHeadÚcreate_act_layerÚConvNormActÚDropPathÚmake_divisibleÚSEModuleÚget_device_dtypeé   )Úbuild_model_with_cfg)Úefficientnet_init_weights)Úfeature_take_indices)Ú
checkpointÚcheckpoint_seq)Úgenerate_default_cfgsÚregister_modelÚRexNet)Ú
norm_layerc                   ó¼   ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 	 d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
                 fˆ fd„Zddedefd„Zdej        dej        fd„Zˆ xZS )ÚLinearBottleneckz‡Linear bottleneck block for ReXNet.

    A mobile inverted residual bottleneck block as used in MobileNetV2 and subsequent models.
    ©r   r   ç      ð?ç        r   ÚswishÚrelu6NÚin_chsÚout_chsÚstrideÚdilationÚ	exp_ratioÚse_ratioÚch_divÚ	act_layerÚdw_act_layerÚ	drop_pathc           
      ó`  •— ||dœ}t          ¦   «                              ¦   «          |dk    o|d         |d         k    o||k    | _        || _        || _        |dk    r7t          t          ||z  ¦  «        |¬¦  «        }t          ||fd|i|¤Ž| _        n	|}d| _        t          ||fd||d         |d	d
œ|¤Ž| _	        |dk    r3t          |fdt          t          ||z  ¦  «        |¦  «        i|¤Ž| _        nd| _        t          |	¦  «        | _        t          ||dfdd	i|¤Ž| _        |
| _        dS )aþ  Initialize LinearBottleneck.

        Args:
            in_chs: Number of input channels.
            out_chs: Number of output channels.
            stride: Stride for depthwise conv.
            dilation: Dilation rates.
            exp_ratio: Expansion ratio.
            se_ratio: Squeeze-excitation ratio.
            ch_div: Channel divisor.
            act_layer: Activation layer for expansion.
            dw_act_layer: Activation layer for depthwise.
            drop_path: Drop path module.
        ©ÚdeviceÚdtyper   r   r$   ©Údivisorr/   Né   F)Úkernel_sizer*   r+   ÚgroupsÚ	apply_actÚrd_channelsr;   )ÚsuperÚ__init__Úuse_shortcutÚin_channelsÚout_channelsr   Úroundr   Úconv_expÚconv_dwÚ
SEWithNormÚintÚser   Úact_dwÚconv_pwlr1   )Úselfr(   r)   r*   r+   r,   r-   r.   r/   r0   r1   r4   r5   ÚddÚdw_chsÚ	__class__s                  €úb/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/rexnet.pyr>   zLinearBottleneck.__init__(   sy  ø€ ð: ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ" ašKÐ\¨H°Q¬K¸8ÀA¼;Ò,FÐ\È6ÐU\ÒK\ˆÔØ!ˆÔØ#ˆÔà˜Š?ˆ?Ý#¥E¨&°9Ñ*<Ñ$=Ô$=ÀvÐNÑNÔNˆFÝ'¨°ÐRÐRÀ)ÐRÈrÐRÐRˆDŒMˆMàˆFØ ˆDŒMå"ØØð	
ð ØØ˜a”[ØØð	
ð 	
ð ð	
ð 	
ˆŒð �aŠ<ˆ<Ý  ÐjÐjµ^ÅCÈÐQYÑHYÑDZÔDZÐ\bÑ5cÔ5cÐjÐgiÐjÐjˆDŒGˆGàˆDŒGÝ& |Ñ4Ô4ˆŒå# F¨G°QÐNÐNÀ%ÐNÈ2ÐNÐNˆŒØ"ˆŒˆˆó    FÚexpÚreturnc                 ó,   — |r| j         j        n| j        S )zŸGet feature channel count.

        Args:
            exp: Return expanded channels if True.

        Returns:
            Number of feature channels.
        )rD   rA   )rJ   rP   s     rN   Úfeat_channelszLinearBottleneck.feat_channelse   s   € ð -0ÐFˆtŒ|Ô(Ð(°TÔ5FÐFrO   Úxc                 ó¸  — |}| j         �|                       |¦  «        }|                      |¦  «        }| j        �|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        rY| j        �|                      |¦  «        }t          j        |dd…d| j	        …f         |z   |dd…| j	        d…f         gd¬¦  «        }|S )zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        Nr   r   )Údim)
rC   rD   rG   rH   rI   r?   r1   ÚtorchÚcatr@   )rJ   rT   Úshortcuts      rN   ÚforwardzLinearBottleneck.forwardp   sÞ   € ð ˆØŒ=Ð$Ø—’˜aÑ Ô ˆAØ�LŠL˜‰OŒOˆØŒ7ÐØ—’˜‘
”
ˆAØ�KŠK˜‰NŒNˆØ�MŠM˜!ÑÔˆØÔð 	aØŒ~Ð)Ø—N’N 1Ñ%Ô%�Ý”	˜1˜Q˜Q˜Q  $Ô"2Ð 2Ð2Ô3°hÑ>ÀÀ!À!À!ÀTÔEUÐEVÐEVÐBVÔ@WÐXÐ^_Ð`Ñ`Ô`ˆAØˆrO   )	r#   r$   r%   r   r&   r'   NNN©F)Ú__name__Ú
__module__Ú__qualname__Ú__doc__rF   r
   ÚfloatÚstrr	   ÚnnÚModuler>   ÚboolrS   rW   ÚTensorrZ   Ú__classcell__©rM   s   @rN   r"   r"   "   s/  ø€ € € € € ðð ð )/Ø"Ø ØØ$Ø 'Ø-1ØØð;#ð ;#àð;#ð ð;#ð ð	;#ð
 ˜C ˜H”oð;#ð ð;#ð ð;#ð ð;#ð ð;#ð ð;#ð   ¤	Ô*ð;#ð ;#ð ;#ð ;#ð ;#ð ;#ðz	Gð 	G ð 	G°#ð 	Gð 	Gð 	Gð 	Gð˜œð ¨%¬,ð ð ð ð ð ð ð ð rO   r"   r$   é   é´   r%   Ú
width_multÚ
depth_multÚinitial_chsÚ	final_chsr-   r.   rQ   c           	      ót  ‡‡— g d¢Šg d¢}ˆfd„‰D ¦   «         Št          ˆfd„t          |¦  «        D ¦   «         g ¦  «        }dg‰d         z  dgt          ‰dd…         ¦  «        z  z   }t          ‰dd…         ¦  «        d	z  }| d
k     r|| z  n|}	g }
t          |d	z  ¦  «        D ]D}|
                     t	          t          |	| z  ¦  «        |¬¦  «        ¦  «         |	||d	z  d
z  z  z  }	ŒEdg‰d         ‰d         z   z  |gt          ‰dd…         ¦  «        z  z   }t          t          |
|||¦  «        ¦  «        S )as  Generate ReXNet block configuration.

    Args:
        width_mult: Width multiplier.
        depth_mult: Depth multiplier.
        initial_chs: Initial channel count.
        final_chs: Final channel count.
        se_ratio: Squeeze-excitation ratio.
        ch_div: Channel divisor.

    Returns:
        List of tuples (out_channels, exp_ratio, stride, se_ratio).
    )r   é   ro   r8   r8   é   )r   ro   ro   ro   r   ro   c                 ó4   •— g | ]}t          |‰z  ¦  «        ‘ŒS © r   )Ú.0Úelementrk   s     €rN   ú
<listcomp>z_block_cfg.<locals>.<listcomp>Ÿ   s&   ø€ Ð?Ð?Ð?¨W�d�7˜ZÑ'Ñ(Ô(Ð?Ð?Ð?rO   c                 ó<   •— g | ]\  }}|gd g‰|         d z
  z  z   ‘ŒS )r   rr   )rs   Úidxrt   Úlayerss      €rN   ru   z_block_cfg.<locals>.<listcomp>    s3   ø€ Ð^Ð^Ð^¹<¸3À�G�9 ˜s f¨S¤k°A¡oÑ6Ñ6Ð^Ð^Ð^rO   r   r   é   Nr8   r$   r6   r%   ro   )ÚsumÚ	enumerateÚrangeÚappendr   rB   ÚlistÚzip)rj   rk   rl   rm   r-   r.   ÚstridesÚ
exp_ratiosÚdepthÚbase_chsÚout_chs_listÚiÚ	se_ratiosrx   s    `           @rN   Ú
_block_cfgr‡   ˆ   sw  øø€ ð*  ÐÐ€FØ Ð Ð €GØ?Ð?Ð?Ð?¸Ð?Ñ?Ô?€FÝÐ^Ð^Ð^Ð^Í9ÐU\ÑK]ÔK]Ð^Ñ^Ô^Ð`bÑcÔc€GØ��v˜a”y‘ A 3­¨V°A°B°B¬Z©¬Ñ#8Ñ8€JÝ��q�q�q”	‰NŒN˜QÑ€EØ+5¸Ò+;Ð+;ˆ{˜ZÑ'Ð'À€Hð €LÝ�5˜A‘:ÑÔð 3ð 3ˆØ×Ò�N­5°¸JÑ1FÑ+GÔ+GÐQWÐXÑXÔXÑYÔYÐYØ�I ¨!¡¨cÑ!1Ñ2Ñ2ˆˆà�˜˜qœ	 F¨1¤IÑ-Ñ.°(°½cÀ&ÈÈÈÄ*¹o¼oÑ1MÑM€Iå•�L *¨g°yÑAÔAÑBÔBÐBrO   é    r&   r'   Ú	block_cfgÚprev_chsÚoutput_strider/   r0   Údrop_path_ratec
                 ó´  — ||	dœ}
|g}g }d}d}g }t          | ¦  «        }t          | ¦  «        D ]À\  }\  }}}}|}|dk    r9|dk    rdnd|dz
  › �}|t          |d         ||¬¦  «        gz  }||k    r||z  }d}||z  |dz
  z  }|d	k    rt          |¦  «        nd
}|                     t          d||||||f|||||dœ
|
¤Ž¦  «         ||z  }|}|}||d                              ¦   «         gz  }ŒÁt          d|z  |¬¦  «        }|t          |d         |dt          |¦  «        dz
  › �¬¦  «        gz  }|                     t          ||fd|i|
¤Ž¦  «         ||fS )aÎ  Build ReXNet blocks from configuration.

    Args:
        block_cfg: Block configuration list.
        prev_chs: Previous channel count.
        width_mult: Width multiplier.
        ch_div: Channel divisor.
        output_stride: Target output stride.
        act_layer: Activation layer name.
        dw_act_layer: Depthwise activation layer name.
        drop_path_rate: Drop path rate.

    Returns:
        Tuple of (features list, feature_info list).
    r3   ro   r   r   Ústemz	features.éÿÿÿÿ)Únum_chsÚ	reductionÚmoduler%   N)
r(   r)   r,   r*   r+   r-   r.   r/   r0   r1   i   r6   r/   rr   )	Úlenr{   Údictr   r}   r"   rS   r   r   )r‰   rŠ   rj   r.   r‹   r/   r0   rŒ   r4   r5   rK   Úfeat_chsÚfeature_infoÚcurr_strider+   ÚfeaturesÚ
num_blocksÚ	block_idxÚchsr,   r*   r-   Únext_dilationÚfnameÚ	block_dprr1   Úpen_chss                              rN   Ú_build_blocksr    °   s  € ð6  UÐ	+Ð	+€BØˆz€HØ€LØ€KØ€HØ€HÝ�Y‘”€JÝ9BÀ9Ñ9MÔ9Mð 3ð 3Ñ5ˆ	Ñ5�C˜ F¨HØ ˆØ�AŠ:ˆ:Ø'¨1šn˜n�F�FÐ2M¸iÈ!¹mÐ2MÐ2MˆEØ�T¨(°2¬,À+ÐV[Ð\Ñ\Ô\Ð]Ñ]ˆLØ˜mÒ+Ð+Ø (¨6Ñ 1�Ø�Ø" YÑ.°*¸q±.ÑAˆ	Ø+4°rª>¨>•H˜YÑ'Ô'Ð'¸tˆ	Ø�ŠÕ(ð 
ØØØØØ Ð.ØØØØ%Øð
ð 
ð ð
ð 
ñ 	ô 	ð 	ð 	�vÑˆØ ˆØˆØ�X˜b”\×/Ò/Ñ1Ô1Ð2Ñ2ˆˆÝ˜T JÑ.¸Ð?Ñ?Ô?€GØ•T (¨2¤,¸+ÐNmÕZ]Ð^fÑZgÔZgÐjkÑZkÐNmÐNmÐnÑnÔnÐoÑo€LØ‡O‚O•K ¨'ÐMÐM¸YÐMÈ"ÐMÐMÑNÔNÐNØ�\Ð!Ð!rO   c                   óÈ  ‡ — e Zd ZdZ	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 	 d7d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defˆ fd„Zej	        j
        d8d ed!eeef         fd"„¦   «         Zej	        j
        d9d$ed!dfd%„¦   «         Zej	        j
        d!ej        fd&„¦   «         Zd:dedee         d!dfd'„Z	 	 	 	 	 d;d)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	 	 	 d<d*eeee         f         d0ed1ed!ee         fd2„Zd)ej        d!ej        fd3„Zd8d)ej        d4ed!ej        fd5„Zd)ej        d!ej        fd6„Zˆ xZS )=r   z¦ReXNet model architecture.

    Based on `ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network`
    - https://arxiv.org/abs/2007.00992
    r8   éè  Úavgrˆ   rh   ri   r$   çUUUUUUµ?r   r&   r'   çš™™™™™É?r%   NÚin_chansÚnum_classesÚglobal_poolr‹   rl   rm   rj   rk   r-   r.   r/   r0   Ú	drop_raterŒ   c           
      ó*  •— t          ¦   «                              ¦   «          ||dœ}|| _        || _        || _        d| _        |dv sJ ‚|dk     rd|z  nd}t          t          ||z  ¦  «        |
¬¦  «        }t          ||dfd|d	œ|¤Ž| _	        t          |||||	|
¦  «        }t          ||||
||||fi |¤Ž\  }| _        |d
         j        x| _        | _        t!          j        |Ž | _        t'          | j        |||fi |¤Ž| _        t+          | ¦  «         dS )a°  Initialize ReXNet.

        Args:
            in_chans: Number of input channels.
            num_classes: Number of classes for classification.
            global_pool: Global pooling type.
            output_stride: Output stride.
            initial_chs: Initial channel count.
            final_chs: Final channel count.
            width_mult: Width multiplier.
            depth_mult: Depth multiplier.
            se_ratio: Squeeze-excitation ratio.
            ch_div: Channel divisor.
            act_layer: Activation layer name.
            dw_act_layer: Depthwise activation layer name.
            drop_rate: Dropout rate.
            drop_path_rate: Drop path rate.
        r3   F)rˆ   rh   é   r$   rˆ   r6   r8   ro   )r*   r/   r�   N)r=   r>   r§   r¦   r©   Úgrad_checkpointingr   rB   r   rŽ   r‡   r    r–   rA   Únum_featuresÚhead_hidden_sizerb   Ú
Sequentialr˜   r   Úheadr   )rJ   r¦   r§   r¨   r‹   rl   rm   rj   rk   r-   r.   r/   r0   r©   rŒ   r4   r5   rK   Ústem_base_chsÚstem_chsr‰   r˜   rM   s                         €rN   r>   zRexNet.__init__ú   sd  ø€ õJ 	‰Œ×ÒÑÔÐØ¨Ð/Ð/ˆØ&ˆÔØ ˆŒØ"ˆŒØ"'ˆÔà Ð+Ð+Ð+Ð+Ø+5¸Ò+;Ð+;˜˜Z™˜ÀˆÝ!¥%¨¸
Ñ(BÑ"CÔ"CÈVÐTÑTÔTˆÝ ¨(°AÐ[¸aÈ9Ð[Ð[ÐXZÐ[Ð[ˆŒ	å˜z¨:°{ÀIÈxÐY_Ñ`Ô`ˆ	Ý&3ØØØØØØØØð
'
ð 
'
ð ð
'
ð 
'
Ñ#ˆ�$Ô#ð 5=¸R´LÔ4MÐMˆÔ˜DÔ1Ýœ xÐ0ˆŒå" 4Ô#4°kÀ;ÐPYÐ`Ð`Ð]_Ð`Ð`ˆŒ	å! $Ñ'Ô'Ð'Ð'Ð'rO   FÚcoarserQ   c                 ó(   — t          dd¬¦  «        }|S )z¯Group matcher for parameter groups.

        Args:
            coarse: Whether to use coarse grouping.

        Returns:
            Dictionary of grouped parameters.
        z^stemz^features\.(\d+))rŽ   Úblocks)r”   )rJ   r³   Úmatchers      rN   Úgroup_matcherzRexNet.group_matcher>  s%   € õ ØØ&ð
ñ 
ô 
ˆð ˆrO   TÚenablec                 ó   — || _         dS )zEnable or disable gradient checkpointing.

        Args:
            enable: Whether to enable gradient checkpointing.
        N)r¬   )rJ   r¸   s     rN   Úset_grad_checkpointingzRexNet.set_grad_checkpointingN  s   € ð #)ˆÔÐÐrO   c                 ó   — | j         j        S )zTGet the classifier module.

        Returns:
            Classifier module.
        )r°   Úfc)rJ   s    rN   Úget_classifierzRexNet.get_classifierW  s   € ð ŒyŒ|ÐrO   c                 ó  — t          | ||¬¦  «        }|| _        |€|�U|�|n| j        j        j        }t          | j        ||| j        fi |¤Ž| _        | j                             | j	        ¦  «         dS  | j        j
        ||fi |¤Ž dS )zšReset the classifier.

        Args:
            num_classes: Number of classes for new classifier.
            global_pool: Global pooling type.
        r3   N)r   r§   r°   r¨   Ú	pool_typer   r­   r©   ÚtrainÚtrainingÚreset)rJ   r§   r¨   r4   r5   rK   r¿   s          rN   Úreset_classifierzRexNet.reset_classifier`  s¦   € õ ˜d¨6¸Ð?Ñ?Ô?ˆØ&ˆÔØÐ Ð!2Ø'2Ð'>˜˜ÀDÄIÔDYÔDcˆIÝ& tÔ'8¸+ÀyÐRVÔR`ÐgÐgÐdfÐgÐgˆDŒIØŒI�OŠO˜DœMÑ*Ô*Ð*Ð*Ð*àˆDŒIŒO˜K¨Ð;Ð;¸Ð;Ð;Ð;Ð;Ð;rO   ÚNCHWrT   ÚindicesÚnormÚ
stop_earlyÚ
output_fmtÚintermediates_onlyc                 ó>  ‡— |dv s
J d¦   «         ‚g }d„ | j         D ¦   «         Št          t          ‰¦  «        |¦  «        \  }}	ˆfd„|D ¦   «         }‰|	         }	|                      |¦  «        }t          j                             ¦   «         s|s| j        }
n| j        d|	dz   …         }
t          |
¦  «        D ]_\  }}| j	        r/t          j                             ¦   «         st          ||¦  «        }n ||¦  «        }||v r|                     |¦  «         Œ`|r|S ||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:

        )rÄ   zOutput shape must be NCHW.c                 ój   — g | ]0}t          |d                               d¦  «        d         ¦  «        ‘Œ1S ©r’   ú.r�   ©rF   Úsplit©rs   Úinfos     rN   ru   z0RexNet.forward_intermediates.<locals>.<listcomp>‡  ó7   € ÐWÐWÐW¸T•c˜$˜xœ.×.Ò.¨sÑ3Ô3°BÔ7Ñ8Ô8ÐWÐWÐWrO   c                 ó    •— g | ]
}‰|         ‘ŒS rr   rr   )rs   r…   Ú
stage_endss     €rN   ru   z0RexNet.forward_intermediates.<locals>.<listcomp>‰  s   ø€ Ð<Ð<Ð<¨!˜
 1œÐ<Ð<Ð<rO   Nr   )r–   r   r“   rŽ   rW   ÚjitÚis_scriptingr˜   r{   r¬   r   r}   )rJ   rT   rÅ   rÆ   rÇ   rÈ   rÉ   ÚintermediatesÚtake_indicesÚ	max_indexÚstagesÚfeat_idxÚstagerÔ   s                @rN   Úforward_intermediateszRexNet.forward_intermediatesp  sT  ø€ ð* ˜YÐ&Ð&Ð&Ð(DÑ&Ô&Ð&ØˆØWÐWÀTÔEVÐWÑWÔWˆ
Ý"6µs¸:±´ÈÑ"PÔ"PÑˆ�iØ<Ð<Ð<Ð<¨|Ð<Ñ<Ô<ˆØ˜yÔ)ˆ	ð �IŠI�a‰LŒLˆÝŒ9×!Ò!Ñ#Ô#ð 	3¨:ð 	3Ø”]ˆFˆFà”] > I°¡M >Ô2ˆFå(¨Ñ0Ô0ð 	(ð 	(‰OˆH�eØÔ&ð ­u¬y×/EÒ/EÑ/GÔ/Gð Ý˜u aÑ(Ô(��à�E˜!‘H”H�Ø˜<Ð'Ð'Ø×$Ò$ QÑ'Ô'Ð'øàð 	!Ø Ð à�-ÐÐrO   Ú
prune_normÚ
prune_headc                 óÖ   — d„ | j         D ¦   «         }t          t          |¦  «        |¦  «        \  }}||         }| j        d|dz   …         | _        |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.
        c                 ój   — g | ]0}t          |d                               d¦  «        d         ¦  «        ‘Œ1S rÌ   rÎ   rÐ   s     rN   ru   z4RexNet.prune_intermediate_layers.<locals>.<listcomp>°  rÒ   rO   Nr   r   Ú )r–   r   r“   r˜   rÃ   )rJ   rÅ   rÞ   rß   rÔ   rØ   rÙ   s          rN   Úprune_intermediate_layersz RexNet.prune_intermediate_layers   sy   € ð  XÐWÀTÔEVÐWÑWÔWˆ
Ý"6µs¸:±´ÈÑ"PÔ"PÑˆ�iØ˜yÔ)ˆ	Øœ n y°1¡} nÔ5ˆŒØð 	)Ø×!Ò! ! RÑ(Ô(Ð(ØÐrO   c                 óÐ   — |                       |¦  «        }| j        r4t          j                             ¦   «         st          | j        |¦  «        }n|                      |¦  «        }|S )z’Forward pass through feature extraction layers.

        Args:
            x: Input tensor.

        Returns:
            Feature tensor.
        )rŽ   r¬   rW   rÕ   rÖ   r   r˜   ©rJ   rT   s     rN   Úforward_featureszRexNet.forward_features¸  s\   € ð �IŠI�a‰LŒLˆØÔ"ð 	!­5¬9×+AÒ+AÑ+CÔ+Cð 	!Ý˜tœ}¨aÑ0Ô0ˆAˆAà—’˜aÑ Ô ˆAØˆrO   Ú
pre_logitsc                 ó^   — |r|                       ||¬¦  «        n|                       |¦  «        S )zÕForward pass through head.

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

        Returns:
            Classification logits or features.
        )rç   )r°   )rJ   rT   rç   s      rN   Úforward_headzRexNet.forward_headÈ  s0   € ð 7AÐRˆt�yŠy˜ zˆyÑ2Ô2Ð2ÀdÇiÂiÐPQÁlÄlÐRrO   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S )zoForward pass.

        Args:
            x: Input tensor.

        Returns:
            Output logits.
        )ræ   ré   rå   s     rN   rZ   zRexNet.forwardÔ  s/   € ð ×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆrO   )r8   r¢   r£   rˆ   rh   ri   r$   r$   r¤   r   r&   r'   r¥   r%   NNr[   )T)NNN)NFFrÄ   F)r   FT)r\   r]   r^   r_   rF   ra   r`   r>   rW   rÕ   Úignorerd   r   r   r·   rº   rb   rc   r½   r	   rÃ   re   r   r   r
   rÝ   rã   ræ   ré   rZ   rf   rg   s   @rN   r   r   ó   s‚  ø€ € € € € ðð ð Ø#Ø$Ø!#Ø!Ø Ø #Ø #Ø#ØØ$Ø 'Ø"Ø$&ØØð#B(ð B(àðB(ð ðB(ð ð	B(ð
 ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð ðB(ð "ðB(ð B(ð B(ð B(ð B(ð B(ðH „YÔðð  Dð °T¸#¸s¸(´^ð ð ð ñ Ôðð „YÔð)ð )¨Tð )¸Tð )ð )ð )ñ Ôð)ð „YÔð ¤	ð ð ð ñ Ôðð<ð <¨Cð <¸hÀs¼mð <Ðptð <ð <ð <ð <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 ð0 %¤,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 rO   ÚvariantÚ
pretrainedc                 óN   — t          d¬¦  «        }t          t          | |fd|i|¤ŽS )zÓCreate a ReXNet model.

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

    Returns:
        ReXNet model instance.
    T)Úflatten_sequentialÚfeature_cfg)r”   r   r   )rì   rí   Úkwargsrð   s       rN   Ú_create_rexnetrò   â  sH   € õ ¨$Ð/Ñ/Ô/€KÝÝØØðð ð  ð	ð
 ðð ð rO   râ   Úurlc                 ó6   — | 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.
    r¢   )r8   éà   rõ   )é   rö   g      ì?Úbicubicz	stem.convzhead.fcÚmit)ró   r§   Ú
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚmeanÚstdÚ
first_convÚ
classifierÚlicenser   )ró   rñ   s     rN   Ú_cfgr  ÷  s:   € ð  4°}ÐSYØ¨IÝ%Õ.BØ!°Øðð ð
 #ðð rO   ztimm/)Ú	hf_hub_idgffffffî?)r8   é   r  z
apache-2.0)r  rû   Útest_crop_pctÚtest_input_sizer  i-.  )r  r§   rû   r  r  r  )zrexnet_100.nav_in1kzrexnet_130.nav_in1kzrexnet_150.nav_in1kzrexnet_200.nav_in1kzrexnet_300.nav_in1kzrexnetr_100.untrainedzrexnetr_130.untrainedzrexnetr_150.untrainedzrexnetr_200.sw_in12k_ft_in1kzrexnetr_300.sw_in12k_ft_in1kzrexnetr_200.sw_in12kzrexnetr_300.sw_in12kFc                 ó   — t          d| fi |¤ŽS )zReXNet V1 1.0xÚ
rexnet_100©rò   ©rí   rñ   s     rN   r  r  $  s   € õ ˜,¨
Ð=Ð=°fÐ=Ð=Ð=rO   c                 ó"   — t          d| fddi|¤ŽS )zReXNet V1 1.3xÚ
rexnet_130rj   çÍÌÌÌÌÌô?r	  r
  s     rN   r  r  *  ó!   € õ ˜,¨
ÐMÐM¸sÐMÀfÐMÐMÐMrO   c                 ó"   — t          d| fddi|¤ŽS )zReXNet V1 1.5xÚ
rexnet_150rj   ç      ø?r	  r
  s     rN   r  r  0  r  rO   c                 ó"   — t          d| fddi|¤ŽS )zReXNet V1 2.0xÚ
rexnet_200rj   ç       @r	  r
  s     rN   r  r  6  r  rO   c                 ó"   — t          d| fddi|¤ŽS )zReXNet V1 3.0xÚ
rexnet_300rj   ç      @r	  r
  s     rN   r  r  <  r  rO   c                 ó"   — t          d| fddi|¤ŽS )z*ReXNet V1 1.0x w/ rounded (mod 8) channelsÚrexnetr_100r.   r«   r	  r
  s     rN   r  r  B  s!   € õ ˜-¨ÐHÐH¸AÐHÀÐHÐHÐHrO   c                 ó$   — t          d| fdddœ|¤ŽS )z*ReXNet V1 1.3x w/ rounded (mod 8) channelsÚrexnetr_130r  r«   ©rj   r.   r	  r
  s     rN   r  r  H  ó$   € õ ˜-¨ÐXÀÈAÐXÐXÐQWÐXÐXÐXrO   c                 ó$   — t          d| fdddœ|¤ŽS )z*ReXNet V1 1.5x w/ rounded (mod 8) channelsÚrexnetr_150r  r«   r  r	  r
  s     rN   r  r  N  r  rO   c                 ó$   — t          d| fdddœ|¤ŽS )z*ReXNet V1 2.0x w/ rounded (mod 8) channelsÚrexnetr_200r  r«   r  r	  r
  s     rN   r!  r!  T  r  rO   c                 ó$   — t          d| fdddœ|¤ŽS )z+ReXNet V1 3.0x w/ rounded (mod 16) channelsÚrexnetr_300r  rh   r  r	  r
  s     rN   r#  r#  Z  s$   € õ ˜-¨ÐYÀÈBÐYÐYÐRXÐYÐYÐYrO   )r$   r$   rh   ri   r%   r   )r   rˆ   r&   r'   r%   NN)râ   r[   )@r_   Ú	functoolsr   Úmathr   Útypingr   r   r   r	   r
   r   r   rW   Útorch.nnrb   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   Ú_builderr   Ú_efficientnet_builderr   Ú	_featuresr   Ú_manipulater   r   Ú	_registryr   r   Ú__all__ÚBatchNorm2drE   rc   r"   r`   rF   r‡   ra   r    r   rd   rò   r  Údefault_cfgsr  r  r  r  r  r  r  r  r!  r#  rr   rO   rN   ú<module>r2     sÝ  ðð
ð 
ð Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ø @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @à €€€Ø Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ AØ {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ð {Ø *Ð *Ð *Ð *Ð *Ð *Ø <Ð <Ð <Ð <Ð <Ð <Ø +Ð +Ð +Ð +Ð +Ð +Ø 3Ð 3Ð 3Ð 3Ð 3Ð 3Ð 3Ð 3Ø <Ð <Ð <Ð <Ð <Ð <Ð <Ð <àˆ*€ð ˆW�X¨"¬.Ð9Ñ9Ô9€
ðcð cð cð cð c�r”yñ cô cð cðN  ØØØØØð%Cð %CØð%Càð%Cð ð%Cð ð	%Cð
 ð%Cð ð%Cð 
ˆ%��U˜C Ð&Ô
'Ô(ð%Cð %Cð %Cð %CðX ØØ Ø#Ø "ØØð@"ð @"Ø˜˜c 5¨#¨uÐ4Ô5Ô6ð@"àð@"ð ð@"ð ð	@"ð
 ð@"ð ð@"ð ð@"ð ð@"ð ˆ4�”	Œ?˜D  c¨3 h¤Ô0Ð0Ô1ð@"ð @"ð @"ð @"ðFlð lð lð lð lˆRŒYñ lô lð lð^˜Cð ¨Tð Àð ð ð ð ð*ð ˆcð  T¨#¨s¨(¤^ð ð ð ð ð& %Ð$Ø˜4¨'Ð2Ñ2Ô2Ø˜4¨'Ð2Ñ2Ô2Ø˜4¨'Ð2Ñ2Ô2Ø˜4¨'Ð2Ñ2Ô2Ø˜4¨'Ð2Ñ2Ô2Ø!˜T™VœVØ!˜T™VœVØ!˜T™VœVØ$( DØØ S¸-ÐQ]ð%_ñ %_ô %_ð %) DØØ S¸-ÐQ]ð%_ñ %_ô %_ð !˜DØØØ S¸-ÐQ]ð_ñ _ô _ð !˜DØØØ S¸-ÐQ]ð_ñ _ô _ð'&ð &ñ ô €ð4 ð>ð >˜4ð >°fð >ð >ð >ñ „ð>ð
 ðNð N˜4ð N°fð Nð Nð Nñ „ðNð
 ðNð N˜4ð N°fð Nð Nð Nñ „ðNð
 ðNð N˜4ð N°fð Nð Nð Nñ „ðNð
 ðNð N˜4ð N°fð Nð Nð Nñ „ðNð
 ðIð I˜Dð I°vð Ið Ið Iñ „ðIð
 ðYð Y˜Dð Y°vð Yð Yð Yñ „ðYð
 ðYð Y˜Dð Y°vð Yð Yð Yñ „ðYð
 ðYð Y˜Dð Y°vð Yð Yð Yñ „ðYð
 ðZð Z˜Dð Z°vð Zð Zð Zñ „ðZð Zð ZrO   