j= ddlmZddlmZmZmZmZddlZddlm Z ddlm Z ddl m Z ddl mZdd lmZmZmZdd lmZdd lmZmZgd Zd e dede fdZGdde jZGdde jZdeededededef dZdeddZ GddeZ!Gdd eZ"Gd!d"eZ#Gd#d$eZ$eed%e!j%f&dd'd(dee!dededefd)Z&eed%e"j%f&dd'd(dee"dededefd*Z'eed%e#j%f&dd'd(dee#dededefd+Z(eed%e$j%f&dd'd(dee$dededefd,Z)dS)-)partial)AnyCallableListOptionalN)Tensor)ImageClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) ShuffleNetV2ShuffleNet_V2_X0_5_WeightsShuffleNet_V2_X1_0_WeightsShuffleNet_V2_X1_5_WeightsShuffleNet_V2_X2_0_Weightsshufflenet_v2_x0_5shufflenet_v2_x1_0shufflenet_v2_x1_5shufflenet_v2_x2_0xgroupsreturnc|\}}}}||z}||||||}tj|dd}|||||}|S)Nr r )sizeviewtorch transpose contiguous)rr batchsize num_channelsheightwidthchannels_per_groups PC:\Projects\Historical\venv\Lib\site-packages\torchvision/models/shufflenetv2.pychannel_shuffler+s|-.VVXX*I|VU%/ y&"4feDDA 1a  ++--A y,66A HceZdZdedededdffd Ze dd ed ed eded ededejfdZ de de fdZ xZ S)InvertedResidualinpoupstriderNctd|cxkrdksntd||_|dz}|jdkr"||dzkrtd|d|d|d|jdkrt j|||d|jd t j|t j||ddd d t j|t j d |_ nt j|_ t jt j|jdkr|n||ddd d t j|t j d |||d|jd t j|t j||ddd d t j|t j d |_ dS)Nr zillegal stride valuer zInvalid combination of stride z, inp z and oup zB values. If stride == 1 then inp should be equal to oup // 2 << 1. kernel_sizer1paddingrF)r5r1r6biasTinplace) super__init__ ValueErrorr1nn Sequentialdepthwise_conv BatchNorm2dConv2dReLUbranch1branch2)selfr/r0r1branch_features __class__s r*r;zInvertedResidual.__init__,s" V    q    344 4 ( K1  3/Q*>#>#>UUUsUUSUUU  ;??=##C!DKYZ#[[s## #AaQRY^___//%%% DLL=??DL} I ao     N? + + GD ! ! !   aX\Xcmn  o o N? + + IoAaYZaf g g g N? + + GD ! ! !   r,r rFior5r6r7c 8tj|||||||S)N)r7r)r=rA)rHrIr5r1r6r7s r*r?zInvertedResidual.depthwise_convVs#yA{FG$qQQQQr,rcD|jdkrF|dd\}}tj|||fd}n>tj||||fd}t |d}|S)Nr r )dim)r1chunkr"catrDrCr+)rErx1x2outs r*forwardzInvertedResidual.forward\s ;!  WWQAW&&FB)Rb!1!12:::CC)T\\!__dll1oo>AFFFCc1%% r,)r rF) __name__ __module__ __qualname__intr; staticmethodboolr=rAr?rrR __classcell__rGs@r*r.r.+s( C( c( 3( 4( ( ( ( ( ( TZ_RR RR%(R25RDGRSWR RRR\R  F        r,r.c eZdZdefdeedeedededejfddf fd Z d e de fd Z d e de fd Z xZ S) ristages_repeatsstages_out_channels num_classesinverted_residual.rNc 6tt|t|dkrt dt|dkrt d||_d}|jd}t jt j||dddd t j |t j d |_ |}t j ddd |_ |||d dD}t|||jddD]h\}} }|||dg} t| dz D]"} | |||d#t#||t j| |}i|jd}t jt j||dddd t j |t j d |_t j|||_dS)Nr3z2expected stages_repeats as list of 3 positive intsz7expected stages_out_channels as list of 5 positive intsrr r F)r7Tr8r4cg|]}d|S)stage).0rHs r* z)ShuffleNetV2.__init__..s666q{q{{666r,)r r3)r:r;r lenr<_stage_out_channelsr=r>rAr@rBconv1 MaxPool2dmaxpoolziprangeappendsetattrconv5Linearfc) rEr\r]r^r_input_channelsoutput_channels stage_namesnamerepeatsseqrHrGs r*r;zShuffleNetV2.__init__is2 D!!! ~  ! # #QRR R " # #q ( (VWW W#6 215] Inoq!QU K K K N? + + GD ! ! !   )|!QGGG  " " "66I666 .1+~tOghihjhjOk.l.l - - *D'?$$^_aHHIC7Q;'' S S ,,_oqQQRRRR D$ s 3 4 4 4,NN226] Inoq!QU K K K N? + + GD ! ! !   )O[99r,rcZ||}||}||}||}||}||}|ddg}||}|S)Nr r3)rkrmstage2stage3stage4rrmeanrtrErs r* _forward_implzShuffleNetV2._forward_impls JJqMM LLOO KKNN KKNN KKNN JJqMM FFAq6NN GGAJJr,c,||S)N)rrs r*rRzShuffleNetV2.forwards!!!$$$r,)rSrTrUr.rrVrr=Moduler;rrrRrYrZs@r*rrhs  6F .:.:S .:"#Y.: .: $CN3 .:  .:.:.:.:.:.:` v &    %%F%%%%%%%%r,rweightsprogressargskwargsc|)t|dt|jdt|i|}|*|||d|S)Nr^ categoriesT)r check_hash)rrimetarload_state_dictget_state_dict)rrrrmodels r* _shufflenetv2rsp fmSl9S5T5TUUU $ )& ) )E g44hSW4XXYYY Lr,)r r z2https://github.com/ericsun99/Shufflenet-v2-Pytorch)min_sizerrecipec beZdZedeediedddddid d d d  ZeZdS)rzDhttps://download.pytorch.org/models/shufflenetv2_x0.5-f707e7126e.pth crop_sizei ImageNet-1Kg-FN@g9voT@zacc@1zacc@5g{Gz?gT㥛 @VThese weights were trained from scratch to reproduce closely the results of the paper. num_params_metrics_ops _file_size_docsurl transformsrN rSrTrUrrr _COMMON_META IMAGENET1K_V1DEFAULTrdr,r*rrsG R7.#>>>    !##   q    M$GGGr,rc beZdZedeediedddddid d d d  ZeZdS)rzBhttps://download.pytorch.org/models/shufflenetv2_x1-5666bf0f80.pthrri"rgI +WQ@gNbX9V@rg(\?gE!@rrrNrrdr,r*rrsG P7.#>>>    !##   q    M$GGGr,rc feZdZedeeddieddddd d id d d dZeZdS)rzBhttps://download.pytorch.org/models/shufflenetv2_x1_5-3c479a10.pthrr resize_size+https://github.com/pytorch/vision/pull/5906iv5rg9v?R@g/$V@rgl?gw/+@ These weights were trained from scratch by using TorchVision's `new training recipe `_. rrrrrrrNrrdr,r*rrG P7.#3OOO  C!##       M*GGGr,rc feZdZedeeddieddddd d id d d dZeZdS)rzBhttps://download.pytorch.org/models/shufflenetv2_x2_0-8be3c8ee.pthrrrriprgQS@gMb@W@rg-?g+n<@rrrNrrdr,r*rrrr,r pretrained)rT)rrc ^t|}t||gdgdfi|S)a  Constructs a ShuffleNetV2 architecture with 0.5x output channels, as described in `ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design `__. Args: weights (:class:`~torchvision.models.ShuffleNet_V2_X0_5_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.ShuffleNet_V2_X0_5_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.shufflenetv2.ShuffleNetV2`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ShuffleNet_V2_X0_5_Weights :members: rgrg)0`)rverifyrrrrs r*rrs?4)//88G (III7N7N7N Y YRX Y YYr,c ^t|}t||gdgdfi|S)a  Constructs a ShuffleNetV2 architecture with 1.0x output channels, as described in `ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design `__. Args: weights (:class:`~torchvision.models.ShuffleNet_V2_X1_0_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.ShuffleNet_V2_X1_0_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.shufflenetv2.ShuffleNetV2`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ShuffleNet_V2_X1_0_Weights :members: r)rtrir)rrrrs r*rr>?4)//88G (III7P7P7P [ [TZ [ [[r,c ^t|}t||gdgdfi|S)a  Constructs a ShuffleNetV2 architecture with 1.5x output channels, as described in `ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design `__. Args: weights (:class:`~torchvision.models.ShuffleNet_V2_X1_5_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.ShuffleNet_V2_X1_5_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.shufflenetv2.ShuffleNetV2`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ShuffleNet_V2_X1_5_Weights :members: r)ri`ir)rrrrs r*rr]rr,c ^t|}t||gdgdfi|S)a  Constructs a ShuffleNetV2 architecture with 2.0x output channels, as described in `ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design `__. Args: weights (:class:`~torchvision.models.ShuffleNet_V2_X2_0_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.ShuffleNet_V2_X2_0_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.shufflenetv2.ShuffleNetV2`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.ShuffleNet_V2_X2_0_Weights :members: r)riii)rrrrs r*rr|rr,)* functoolsrtypingrrrrr"torch.nnr=rtransforms._presetsr utilsr _apir rr_metar_utilsrr__all__rVr+rr.rrXrrrrrrrrrrrrdr,r*rs000000000000 555555''''''6666666666''''''BBBBBBBB     v  s  v     :::::ry:::z>%>%>%>%>%29>%>%>%B k "    $&B ,,22,0J0X!YZZZ7;dZZZ34ZGKZ^aZZZZ[ZZ:,0J0X!YZZZ7;d\\\34\GK\^a\\\\[Z\:,0J0X!YZZZ7;d\\\34\GK\^a\\\\[Z\:,0J0X!YZZZ7;d\\\34\GK\^a\\\\[Z\\\r,