§
    ·¯¸jMH  ã                   ó   — d Z ddlmZmZmZmZmZmZmZ ddl	Z	ddl
mZ ddlmc mZ ddlmZmZ ddlmZmZmZmZmZmZmZ ddlmZ ddlmZ dd	lmZ dd
l m!Z!m"Z" dgZ# G d„ dej$        ¦  «        Z%dZ& G d„ dej$        ¦  «        Z'dde(de)de'fd„Z*dde(dee(ef         fd„Z+ e! e+d¬¦  «         e+d¬¦  «         e+d¬¦  «         e+d¬¦  «        dœ¦  «        Z,e"dde)de'fd„¦   «         Z-e"dde)de'fd„¦   «         Z.e"dde)de'fd„¦   «         Z/dS ) aR   PP-LCNetV2

Reference:
https://github.com/PaddlePaddle/PaddleClas/blob/release/2.6/docs/en/models/PP-LCNetV2_en.md
The Paddle Implement of PP-LCNetV2 (https://github.com/PaddlePaddle/PaddleClas/blob/release/2.6/ppcls/arch/backbone/legendary_models/pp_lcnet_v2.py)

PP-LCNetV2 is a CPU oriented network built on PP-LCNet (see `lcnet_*` models in mobilenetv3.py). The depthwise
convs of the later stages are re-parameterizable multi-scale branches, SE and shortcuts are used sparingly to
avoid latency penalties on CPU inference.

Adapted from the Paddle implementation for timm by Yonghye Kwon
é    )ÚAnyÚDictÚListÚOptionalÚTupleÚTypeÚUnionN©ÚIMAGENET_DEFAULT_MEANÚIMAGENET_DEFAULT_STD)ÚConvNormActÚSelectAdaptivePool2dÚSqueezeExciteÚcreate_act_layerÚcreate_conv2dÚmake_divisibleÚget_device_dtypeé   )Úbuild_model_with_cfg)Úfeature_take_indices)Úcheckpoint_seq)Úgenerate_default_cfgsÚregister_modelÚLCNetV2c                   óX  ‡ — e Zd ZdZddddddddej        ej        dddfded	ed
edededededede	de	de
ej                 de
ej                 deddfˆ fd„Zdej        dej        fd„Zdd„Zdeej        ej        f         fd„Zededeej        ej        f         fd„¦   «         Zˆ xZS )ÚRepDepthwiseSeparablea*  Depthwise separable block with re-parameterizable multi-scale depthwise branches.

    When `use_rep` is enabled the depthwise conv is trained as a set of parallel conv-bn branches with
    decreasing kernel size (e.g. 5x5, 3x3, 1x1) that are folded into a single depthwise conv for inference.
    r   é   Fg      à?g      Ð?NÚin_chsÚout_chsÚstrideÚdw_kernel_sizeÚsplit_pwÚuse_repÚuse_seÚuse_shortcutÚpw_ratioÚse_ratioÚ	act_layerÚ
norm_layerÚinference_modeÚreturnc           
      ój  •‡‡‡‡— ||dœŠt          ¦   «                              ¦   «          ‰| _        || _        ‰| _        || _        |o‰dk    o‰|k    | _        |rwd| _        |rt          ‰‰f|‰‰ddœ‰¤Ž| _	        n>d| _	        t          j        ˆˆˆˆfd„t          |dd¦  «        D ¦   «         ¦  «        | _        t          |d¬	¦  «        | _        n8d| _        t          ‰‰f|‰‰|‰d
œ‰¤Ž| _	        t          j        ¦   «         | _        |r't#          ‰ft%          ‰|
z  ¦  «        |ddœ‰¤Ž| _        nt          j        ¦   «         | _        |rNt%          ||	z  ¦  «        }t          j        t          ‰|fd|‰dœ‰¤Žt          ||fd|‰dœ‰¤Ž¦  «        | _        dS t          ‰|fd|‰dœ‰¤Ž| _        dS )a£  
        Args:
            in_chs: Number of input channels.
            out_chs: Number of output channels.
            stride: Stride of the depthwise conv.
            dw_kernel_size: Kernel size of the (largest) depthwise conv.
            split_pw: Split the pointwise conv into a squeeze and an expand conv.
            use_rep: Train the depthwise conv as re-parameterizable multi-scale branches.
            use_se: Add a squeeze-excite module after the depthwise conv.
            use_shortcut: Add a residual shortcut when input and output shapes match.
            pw_ratio: Channel ratio of the first pointwise conv when `split_pw` is enabled.
            se_ratio: Channel reduction ratio of the squeeze-excite module.
            act_layer: Type of activation layer.
            norm_layer: Type of normalization layer.
            inference_mode: Instantiate the depthwise conv in its re-parameterized form.
        ©ÚdeviceÚdtyper   NT)Úkernel_sizer    ÚgroupsÚbiasc                 óP   •— g | ]"}|d k    r‰d k    °t          ‰‰f|‰‰d‰dœ‰¤Ž‘Œ#S )r   F)r0   r    r1   Ú	apply_actr)   )r   )Ú.0ÚkÚddr   r)   r    s     €€€€úc/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/timm/models/lcnetv2.pyú
<listcomp>z2RepDepthwiseSeparable.__init__.<locals>.<listcomp>^   sm   ø€ ð 3ð 3ð 3ð ÀAÈÂFÀFÈvÐYZÊ{È{õ  ØØð	ð %&Ø%Ø%Ø"'Ø#-ð	ð 	ð ð	ð 	ð P[È{È{ó    r   éþÿÿÿ©Úinplace)r0   r    r1   r(   r)   Úsigmoid)Úrd_channelsr(   Ú
gate_layer)r0   r(   r)   )ÚsuperÚ__init__r   r   r    r!   r%   Údw_conv_listr   Údw_convÚnnÚ
ModuleListÚranger   Údw_actr   ÚIdentityr   ÚintÚseÚ
SequentialÚpw_conv)Úselfr   r   r    r!   r"   r#   r$   r%   r&   r'   r(   r)   r*   r.   r/   Úmid_chsr7   Ú	__class__s    ` `        `    @€r8   rB   zRepDepthwiseSeparable.__init__%   s‡  øøøøø€ ðD ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØˆŒØˆŒØˆŒØ,ˆÔØ(ÐN¨V°qª[ÐN¸VÀwÒ=NˆÔàð (	(Ø $ˆDÔØð Ý,ØØð ð !/Ø!Ø!Øð ð  ð ð ð  �”�ð  $�”Ý$&¤Mð 3ð 3ð 3ð 3ð 3ð 3ð 3õ  % ^°Q¸Ñ;Ô;ð3ñ 3ô 3ñ %ô %�Ô!õ +¨9¸dÐCÑCÔCˆDŒKˆKà $ˆDÔÝ&ØØð	ð +ØØØ#Ø%ð	ð 	ð ð	ð 	ˆDŒLõ œ+™-œ-ˆDŒKàð 		$Ý#Øðå ¨Ñ 1Ñ2Ô2Ø#Ø$ð	ð ð
 ðð ˆDŒGˆGõ ”k‘m”mˆDŒGàð 	Ý˜' HÑ,Ñ-Ô-ˆGÝœ=Ý˜F GÐm¸ÀiÐ\fÐmÐmÐjlÐmÐmÝ˜G WÐn¸!ÀyÐ]gÐnÐnÐkmÐnÐnñô ˆDŒLˆLˆLõ
 'ØØðð Ø#Ø%ðð ð ðð ˆDŒLˆLˆLr:   Úxc                 óˆ  — |}| j         �)|                      |                       |¦  «        ¦  «        }nY | j        d         |¦  «        }t          | j        ¦  «        D ]\  }}|dk    r| ||¦  «        z   }Œ|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }| j        r||z   }|S )Nr   )rD   rH   rC   Ú	enumeraterK   rM   r%   )rN   rQ   ÚshortcutÚoutÚirD   s         r8   ÚforwardzRepDepthwiseSeparable.forward”   sÅ   € ØˆØŒ<Ð#Ø—’˜DŸLšL¨™OœOÑ,Ô,ˆAˆAà&�$Ô# AÔ& qÑ)Ô)ˆCÝ'¨Ô(9Ñ:Ô:ð +ð +‘
��7Ø�q’5�5Ø  ¨¡
¤
Ñ*�CøØ—’˜CÑ Ô ˆAØ�GŠG�A‰JŒJˆØ�LŠL˜‰OŒOˆØÔð 	Ø�H‘ˆAØˆr:   c           
      ó   — | j         €dS |                      ¦   «         \  }}t          | j        | j        | j        | j        | j        d|j        |j        ¬¦  «        | _        || j        j	        _
        || j        j        _
        | j                              ¦   «         D ]}|                     ¦   «          Œ|                      d¦  «         d| _         dS )zEFold the multi-scale depthwise branches into a single depthwise conv.NT)r0   r    r1   r2   r.   r/   rC   )rC   Ú_get_kernel_biasr   r   r!   r    r.   r/   rD   ÚweightÚdatar2   Ú
parametersÚdetach_Ú__delattr__)rN   Úkernelr2   Úparams       r8   Úreparameterizez$RepDepthwiseSeparable.reparameterize¤   sÏ   € àÔÐ$ØˆFà×,Ò,Ñ.Ô.‰ˆ�Ý$ØŒKØŒKØÔ+Ø”;Ø”;ØØ”=Ø”,ð	
ñ 	
ô 	
ˆŒð $*ˆŒÔÔ Ø!%ˆŒÔÔð Ô&×1Ò1Ñ3Ô3ð 	ð 	ˆEØ�MŠM‰OŒOˆOˆOà×Ò˜Ñ(Ô(Ð(Ø ˆÔÐÐr:   c           	      óÈ   — d}d}| j         D ]S}|                      |¦  «        \  }}| j        |j        d         z
  dz  }|t	          j        |||||g¦  «        z   }||z   }ŒT||fS )zISum the bn folded branch kernels, zero padded to the largest kernel size.r   éÿÿÿÿé   )rC   Ú_fuse_bn_tensorr!   ÚshapeÚFÚpad)rN   Úkernel_finalÚ
bias_finalrD   r_   r2   rh   s          r8   rY   z&RepDepthwiseSeparable._get_kernel_bias¾   s„   € àˆØˆ
ØÔ(ð 	+ð 	+ˆGØ×/Ò/°Ñ8Ô8‰LˆF�DØÔ&¨¬°bÔ)9Ñ9¸aÑ?ˆCØ'­!¬%°¸¸cÀ3ÈÐ8LÑ*MÔ*MÑMˆLØ# dÑ*ˆJˆJØ˜ZÐ'Ð'r:   Úbranchc                 ó  — | j         j        }| j        j        }| j        j        }| j        j        }| j        j        }| j        j        }||z                        ¦   «         }||z                       dddd¦  «        }||z  |||z  |z  z
  fS )z>Fold a conv-bn branch into an equivalent conv weight and bias.rc   r   )	ÚconvrZ   ÚbnÚrunning_meanÚrunning_varr2   ÚepsÚsqrtÚreshape)	rk   r_   ro   rp   ÚgammaÚbetarq   ÚstdÚts	            r8   re   z%RepDepthwiseSeparable._fuse_bn_tensorÉ   s�   € ð ”Ô#ˆØ”yÔ-ˆØ”iÔ+ˆØ”	Ô ˆØŒyŒ~ˆØŒiŒmˆØ˜SÑ ×&Ò&Ñ(Ô(ˆØ�S‰[×!Ò! " a¨¨AÑ.Ô.ˆØ˜‰z˜4 ,°Ñ"6¸Ñ"<Ñ<Ð<Ð<r:   )r+   N)Ú__name__Ú
__module__Ú__qualname__Ú__doc__rE   ÚReLUÚBatchNorm2drJ   ÚboolÚfloatr   ÚModulerB   ÚtorchÚTensorrW   ra   r   rY   Ústaticmethodr   re   Ú__classcell__©rP   s   @r8   r   r      sÄ  ø€ € € € € ðð ð Ø"#Ø"Ø!Ø Ø!&Ø!Ø"Ø)+¬Ø*,¬.Ø#(ØØð!mð màðmð ðmð ð	mð
  ðmð ðmð ðmð ðmð ðmð ðmð ðmð ˜BœI”ðmð ˜RœYœðmð !ðmð" 
ð#mð mð mð mð mð mð^˜œð ¨%¬,ð ð ð ð ð !ð !ð !ð !ð4	( %¨¬°e´lÐ(BÔ"Cð 	(ð 	(ð 	(ð 	(ð ð
= ð 
=°°e´lÀEÄLÐ6PÔ0Qð 
=ð 
=ð 
=ñ „\ð
=ð 
=ð 
=ð 
=ð 
=r:   r   ))é@   r   FFFF)é€   r   FFFF)é   é   TTTF)i   r‰   FTFTc                   ó  ‡ — e Zd ZdZddddddej        ej        dd	d
d
fdedee	df         de	de	de
dee	         deej                 deej                 dededd
fˆ 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
PP-LCNetV2ç      ð?©rd   rd   é   rd   r   éè  Úavgi   ç        FNÚscaleÚdepths.Úin_chansÚnum_classesÚglobal_poolÚhead_hidden_sizer(   r)   Ú	drop_rater*   r+   c                 ó,  •— ||dœ}t          ¦   «                              ¦   «          || _        |	| _        d| _        t          d|||
dœ|¤Ž}t          d|z  ¦  «        }t          d|z  ¦  «        }t          j        t          ||fdd||dœ|¤Žt          ||fd	dd
œ|¤Ž¦  «        | _        g }g | _        d}t          t          ¦  «        D ]™\  }\  }}}}}}t          |dz  |z  ¦  «        }g }t          ||         ¦  «        D ]%}|t          d|||dk    rdnd	|||||dœ|¤Žgz  }|}Œ&|t          j        |Ž gz  }|dz  }| xj        t          ||d|› �¬¦  «        gz  c_        Œšt          j        |Ž | _        || _        |p| j        | _        t'          |¬¦  «        | _        |r7t+          | j        | j        d	fddi|¤Ž| _        t/          |d¬¦  «        | _        n0t          j        ¦   «         | _        t          j        ¦   «         | _        |rt          j        d	¦  «        nt          j        ¦   «         | _        |dk    rt          j        | j        |fi |¤Žnt          j        ¦   «         | _        |                      ¦   «         D �]}t?          |t          j         ¦  «        rSt          j!         "                    |j#        dd¬¦  «         |j$        �$t          j!         %                    |j$        ¦  «         Œpt?          |t          j&        ¦  «        rIt          j!         '                    |j#        ¦  «         t          j!         %                    |j$        ¦  «         ŒÓt?          |t          j        ¦  «        r$t          j!         %                    |j$        ¦  «         �ŒdS )a”  
        Args:
            scale: Channel multiplier.
            depths: Number of blocks per stage.
            in_chans: Number of input image channels.
            num_classes: Number of classes for the classification head.
            global_pool: Type of pooling to use for global pooling features of the FC head.
            head_hidden_size: Number of channels of the pre-logits conv, disabled if None.
            act_layer: Type of activation layer.
            norm_layer: Type of normalization layer.
            drop_rate: Dropout rate.
            inference_mode: Instantiate the re-parameterizable blocks in their re-parameterized form.
        r-   F)r(   r)   r*   é    r†   r   rd   )r0   r    r(   r)   r   )r    r!   r   )r   r   r    r!   r"   r#   r$   r%   zstages.)Únum_chsÚ	reductionÚmodule©Ú	pool_typer2   Tr<   Úfan_outÚrelu)ÚmodeÚnonlinearityN© )(rA   rB   r”   r—   Úgrad_checkpointingÚdictr   rE   rL   r   r   ÚstemÚfeature_inforS   Ú
_STAGE_CFGrG   ÚstagesÚnum_featuresr–   r   r•   r   Ú	conv_headr   Úact2rI   ÚFlattenÚflattenÚLinearÚ
classifierÚmodulesÚ
isinstanceÚConv2dÚinitÚkaiming_normal_rZ   r2   Úzeros_r}   Úones_)rN   r‘   r’   r“   r”   r•   r–   r(   r)   r—   r*   r.   r/   r7   Úblock_kwargsÚstem_chsÚprev_chsr©   r›   Ú	stage_idxÚchsr!   r"   r#   r$   r%   r   ÚblocksÚ	block_idxÚmrP   s                                 €r8   rB   zLCNetV2.__init__ã   sÜ  ø€ ð8 ¨Ð/Ð/ˆÝ‰Œ×ÒÑÔÐØ&ˆÔØ"ˆŒØ"'ˆÔåÐl i¸JÐWeÐlÐlÐikÐlÐlˆÝ! " u¡*Ñ-Ô-ˆÝ! " u¡*Ñ-Ô-ˆÝ”MÝØØðð ØØ#Ø%ðð ð ðð õ " (¨HÐa¸QÈqÐaÐaÐT`ÐaÐañ
ô 
ˆŒ	ð ˆØˆÔØˆ	ÝYbÕcmÑYnÔYnð 	lð 	lÑUˆIÑU˜˜^¨X°wÀÈÝ$ S¨1¡W¨u¡_Ñ5Ô5ˆGØˆFÝ" 6¨)Ô#4Ñ5Ô5ð #ð #�	ØÕ0ð 
Ø#Ø#Ø )¨Q¢ ˜1˜1°AØ#1Ø%Ø#Ø!Ø!-ð
ð 
ð #ð
ð 
ð 
ñ 
�ð #��Ø•r”} fÐ-Ð.Ñ.ˆFØ˜‰NˆIØÐÔ¥$¨wÀ)ÐTiÐ^gÐTiÐTiÐ"jÑ"jÔ"jÐ!kÑkÐÔÐÝ”m VÐ,ˆŒØ$ˆÔð !1Ð E°DÔ4EˆÔÝ/¸+ÐFÑFÔFˆÔØð 	&Ý*¨4Ô+<¸dÔ>SÐUVÐiÐiÐ]bÐiÐfhÐiÐiˆDŒNÝ(¨¸DÐAÑAÔAˆDŒIˆIåœ[™]œ]ˆDŒNÝœ™œˆDŒIØ(3ÐF•r”z !‘}”}�}½¼¹¼ˆŒØQ\Ð_`ÒQ`ÐQ`�"œ) DÔ$9¸;ÐMÐMÈ"ÐMÐMÐMÕfhÔfqÑfsÔfsˆŒà—’‘”ð 		'ñ 		'ˆAÝ˜!�RœYÑ'Ô'ð 'Ý”×'Ò'¨¬°yÈvÐ'ÑVÔVÐVØ”6Ð%Ý”G—N’N 1¤6Ñ*Ô*Ð*øÝ˜A�rœ~Ñ.Ô.ð 'Ý”—’˜aœhÑ'Ô'Ð'Ý”—’˜qœvÑ&Ô&Ð&Ð&Ý˜A�rœyÑ)Ô)ð 'Ý”—’˜qœvÑ&Ô&Ð&ùð		'ð 		'r:   Úcoarsec                 ó,   — t          d|rdnd¬¦  «        S )Nz^stemz^stages\.(\d+)z^stages\.(\d+)\.(\d+))r¦   r½   )r¥   )rN   rÀ   s     r8   Úgroup_matcherzLCNetV2.group_matcherE  s+   € åØØ(.ÐLÐ$Ð$Ð4Lð
ñ 
ô 
ð 	
r:   TÚenablec                 ó   — || _         d S ©N)r¤   )rN   rÃ   s     r8   Úset_grad_checkpointingzLCNetV2.set_grad_checkpointingL  s   € à"(ˆÔÐÐr:   c                 ó   — | j         S rÅ   )r°   )rN   s    r8   Úget_classifierzLCNetV2.get_classifierP  s
   € àŒÐr:   c                 óè  — t          | ||¬¦  «        }|| _        |��t          |¬¦  «        | _        | j                             | j        ¦  «         |rt          j        d¦  «        nt          j        ¦   «         | _	        | j	                             | j        ¦  «         |dk    rt          j
        | j        |fi |¤Žnt          j        ¦   «         | _        | j                             | j        ¦  «         d S )Nr-   r�   r   r   )r   r”   r   r•   ÚtrainÚtrainingrE   r­   rI   r®   r¯   r–   r°   )rN   r”   r•   r.   r/   r7   s         r8   Úreset_classifierzLCNetV2.reset_classifierT  sê   € õ ˜d¨6¸Ð?Ñ?Ô?ˆØ&ˆÔØÐ"Ý3¸kÐJÑJÔJˆDÔØÔ×"Ò" 4¤=Ñ1Ô1Ð1Ø,7ÐJ�2œ: a™=œ=˜=½R¼[¹]¼]ˆDŒLØŒL×Ò˜tœ}Ñ-Ô-Ð-à9DÀqº¸õ œ)ØÔ! ;ð6ð 6Ø24ð6ð 6ð 6ÝNPÌkÉmÌmð 	ŒàŒ×Ò˜dœmÑ,Ô,Ð,Ð,Ð,r:   ÚNCHWrQ   ÚindicesÚnormÚ
stop_earlyÚ
output_fmtÚintermediates_onlyc                 óŒ  — |dv s
J d¦   «         ‚g }t          t          | j        ¦  «        |¦  «        \  }}	|                      |¦  «        }t          j                             ¦   «         s|s| j        }
n| j        d|	dz   …         }
t          |
¦  «        D ])\  }} ||¦  «        }||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:

        )rÍ   zOutput shape must be NCHW.Nr   )	r   Úlenr©   r¦   r�   ÚjitÚis_scriptingrS   Úappend)rN   rQ   rÎ   rÏ   rÐ   rÑ   rÒ   ÚintermediatesÚtake_indicesÚ	max_indexr©   Úfeat_idxÚstages                r8   Úforward_intermediateszLCNetV2.forward_intermediatesf  sí   € ð* ˜YÐ&Ð&Ð&Ð(DÑ&Ô&Ð&ØˆÝ"6µs¸4¼;Ñ7GÔ7GÈÑ"QÔ"QÑˆ�ið �IŠI�a‰LŒLˆÝŒ9×!Ò!Ñ#Ô#ð 	1¨:ð 	1Ø”[ˆFˆFà”[  )¨a¡- Ô0ˆFå(¨Ñ0Ô0ð 	(ð 	(‰OˆH�eØ��a‘”ˆAØ˜<Ð'Ð'Ø×$Ò$ QÑ'Ô'Ð'øàð 	!Ø Ð à�-ÐÐr:   r   Ú
prune_normÚ
prune_headc                 ó®   — t          t          | j        ¦  «        |¦  «        \  }}| j        d|dz   …         | _        |r|                      dd¦  «         |S )z@ Prune layers not required for specified intermediates.
        Nr   r   Ú )r   rÔ   r©   rÌ   )rN   rÎ   rÞ   rß   rÙ   rÚ   s         r8   Úprune_intermediate_layersz!LCNetV2.prune_intermediate_layers�  s\   € õ #7µs¸4¼;Ñ7GÔ7GÈÑ"QÔ"QÑˆ�iØ”k . 9¨q¡= .Ô1ˆŒØð 	)Ø×!Ò! ! RÑ(Ô(Ð(ØÐr:   c                 óÔ   — |                       |¦  «        }| j        r6t          j                             ¦   «         st          | j        |d¬¦  «        }n|                      |¦  «        }|S )NT)r®   )r¦   r¤   r�   rÕ   rÖ   r   r©   ©rN   rQ   s     r8   Úforward_featureszLCNetV2.forward_featuresž  s]   € Ø�IŠI�a‰LŒLˆØÔ"ð 	­5¬9×+AÒ+AÑ+CÔ+Cð 	Ý˜tœ{¨A°tÐ<Ñ<Ô<ˆAˆAà—’˜A‘”ˆAØˆr:   Ú
pre_logitsc                 ó4  — |                       |¦  «        }|                      |¦  «        }|                      |¦  «        }|                      |¦  «        }|r|S | j        dk    r!t          j        || j        | j        ¬¦  «        }|                      |¦  «        S )Nr�   )ÚprË   )	r•   r«   r¬   r®   r—   rg   ÚdropoutrË   r°   )rN   rQ   ræ   s      r8   Úforward_headzLCNetV2.forward_head¦  sŠ   € Ø×Ò˜QÑÔˆØ�NŠN˜1ÑÔˆØ�IŠI�a‰LŒLˆØ�LŠL˜‰OŒOˆØð 	ØˆHØŒ>˜BÒÐÝ”	˜!˜tœ~¸¼ÐFÑFÔFˆAØ�Š˜qÑ!Ô!Ð!r:   c                 óZ   — |                       |¦  «        }|                      |¦  «        }|S rÅ   )rå   rê   rä   s     r8   rW   zLCNetV2.forward±  s-   € Ø×!Ò! !Ñ$Ô$ˆØ×Ò˜aÑ Ô ˆØˆr:   ©F)T)NNN)NFFrÍ   F)r   FT)"rx   ry   rz   r{   rE   r|   r}   r   r   rJ   Ústrr   r   r€   r~   rB   r�   rÕ   Úignorer   r   rÂ   rÆ   rÈ   rÌ   r‚   r	   r   rÝ   râ   rå   rê   rW   r„   r…   s   @r8   r   r   à   sv  ø€ € € € € ØÐð Ø&2ØØ#Ø$Ø.2Ø)+¬Ø*,¬.Ø!Ø#(ØØð`'ð `'àð`'ð ˜#˜s˜(”Oð`'ð ð	`'ð
 ð`'ð ð`'ð ' sœmð`'ð ˜BœI”ð`'ð ˜RœYœð`'ð ð`'ð !ð`'ð 
ð`'ð `'ð `'ð `'ð `'ð `'ðD „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 *.ØØð-ð -àð-ð " #œð-ð 
ð-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 ( ðX ./Ø$Ø#ð	ð à˜3  S¤	˜>Ô*ðð ðð ð	ð
 
ˆcŒðð ð ð ð %¤,ð °5´<ð ð ð ð ð	"ð 	"˜eœlð 	"¸ð 	"ÈÌð 	"ð 	"ð 	"ð 	"ð˜œð ¨%¬,ð ð ð ð ð ð ð ð r:   FÚvariantÚ
pretrainedr+   c                 óx   — |                      dd¦  «        }t          t          | |fdt          d|¬¦  «        i|¤ŽS )NÚout_indices)r   r   rd   r   Úfeature_cfgT)Úflatten_sequentialrò   )Úpopr   r   r¥   )rï   rð   Úkwargsrò   s       r8   Ú_create_lcnetv2r÷   ·  sV   € Ø—*’*˜]¨LÑ9Ô9€KÝÝØØðð õ ¨D¸kÐJÑJÔJð	ð
 ðð ð r:   rá   Úurlc                 ó8   — | dddddt           t          dddd	d
œ|¥S )NrŽ   )r   éà   rú   )é   rû   g      ì?Úbicubiczstem.0.convr°   z*https://github.com/PaddlePaddle/PaddleClasz
apache-2.0)rø   r”   Ú
input_sizeÚ	pool_sizeÚcrop_pctÚinterpolationÚmeanrv   Ú
first_convr°   Ú
origin_urlÚlicenser
   )rø   rö   s     r8   Ú_cfgr  Â  s;   € àØ¨=ÀvØ¨IÝ%Õ.BØ#°<ØBØð	ð 	ð ð	ð 	r:   ztimm/)Ú	hf_hub_id)zlcnetv2_small.paddle_in1kzlcnetv2_base.ssld_in1kzlcnetv2_base.paddle_in1kzlcnetv2_large.paddle_in1kc           	      óV   — t          dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )Ng      è?)rd   rd   é   rd   ©r‘   r’   Úlcnetv2_smallrð   )r
  ©r¥   r÷   ©rð   rö   Ú
model_argss      r8   r
  r
  ×  ó>   € å˜D¨Ð6Ñ6Ô6€JÝÐ`Ð`°zÐ`ÅTÈ*ÐE_ÐE_ÐX^ÐE_ÐE_Ð`Ð`Ð`r:   c           	      óV   — t          dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )Nr‹   rŒ   r	  Úlcnetv2_baserð   )r  r  r  s      r8   r  r  Ý  s>   € å˜C¨Ð5Ñ5Ô5€JÝÐ_Ð_°jÐ_ÅDÈÐD^ÐD^ÐW]ÐD^ÐD^Ð_Ð_Ð_r:   c           	      óV   — t          dd¬¦  «        }t          dd| it          |fi |¤Ž¤ŽS )Ng      ô?)rd   rd   é   rd   r	  Úlcnetv2_largerð   )r  r  r  s      r8   r  r  ã  r  r:   rì   )rá   )0r{   Útypingr   r   r   r   r   r   r	   r�   Útorch.nnrE   Útorch.nn.functionalÚ
functionalrg   Ú	timm.datar   r   Útimm.layersr   r   r   r   r   r   r   Ú_builderr   Ú	_featuresr   Ú_manipulater   Ú	_registryr   r   Ú__all__r€   r   r¨   r   rí   r~   r÷   r  Údefault_cfgsr
  r  r  r£   r:   r8   ú<module>r      s+  ððð ð AÐ @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @Ð @à €€€Ø Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð Ð Ð Ð à AÐ AÐ AÐ AÐ AÐ AÐ AÐ Að%ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %ð %à *Ð *Ð *Ð *Ð *Ð *Ø +Ð +Ð +Ð +Ð +Ð +Ø 'Ð 'Ð 'Ð 'Ð 'Ð 'Ø <Ð <Ð <Ð <Ð <Ð <Ð <Ð <àˆ+€ðv=ð v=ð v=ð v=ð v=˜BœIñ v=ô v=ð v=ðt€
ðTð Tð Tð Tð TˆbŒiñ Tô Tð Tðnð ˜Sð ¨dð Èð ð ð ð ð
ð 
ˆcð 
 T¨#¨s¨(¤^ð 
ð 
ð 
ð 
ð %Ð$Ø!% °Ð!8Ñ!8Ô!8Ø"˜d¨WÐ5Ñ5Ô5Ø $ ¨wÐ 7Ñ 7Ô 7Ø!% °Ð!8Ñ!8Ô!8ð	&ð &ñ ô €ð ðað a˜dð a¸ð að að añ „ðað
 ð`ð `˜Tð `¸ð `ð `ð `ñ „ð`ð
 ðað a˜dð a¸ð að að añ „ðað að ar:   