j-<ddlmZddlmZmZmZddlZddlmZddlm Z ddl m Z dd l mZmZmZdd lmZdd lmZmZmZdd lmZmZmZdd lmZmZmZmZm Z ddlm!Z!ddl"m#Z#gdZ$Gdde!Z%Gddej&Z'Gddej&Z(Gddej&Z)Gddej*Z+dede,dee-de%fd Z.ed!d"d#Z/Gd$d%eZ0Gd&d'eZ1Gd(d)eZ2dede,dee-de%fd*Z3eed+e0j4fd,e j5f-dd.dde j5d/d0ee0d1e-dee,d2ee-d3ee d4ede%fd5Z6eed+e1j4fd,ej5f-dd.ddej5d/d0ee1d1e-dee,d2ee-d3eed4ede%fd6Z7eed+e2j4fd,ej5f-dd.ddej5d/d0ee2d1e-dee,d2ee-d3eed4ede%fd7Z8dS)8)partial)AnyOptionalSequenceN)nn) functional)SemanticSegmentation)register_modelWeights WeightsEnum)_VOC_CATEGORIES)_ovewrite_value_paramhandle_legacy_interfaceIntermediateLayerGetter)mobilenet_v3_largeMobileNet_V3_Large_Weights MobileNetV3)ResNet resnet101ResNet101_Weightsresnet50ResNet50_Weights)_SimpleSegmentationModel)FCNHead) DeepLabV3DeepLabV3_ResNet50_WeightsDeepLabV3_ResNet101_Weights$DeepLabV3_MobileNet_V3_Large_Weightsdeeplabv3_mobilenet_v3_largedeeplabv3_resnet50deeplabv3_resnet101ceZdZdZdS)ra Implements DeepLabV3 model from `"Rethinking Atrous Convolution for Semantic Image Segmentation" `_. Args: backbone (nn.Module): the network used to compute the features for the model. The backbone should return an OrderedDict[Tensor], with the key being "out" for the last feature map used, and "aux" if an auxiliary classifier is used. classifier (nn.Module): module that takes the "out" element returned from the backbone and returns a dense prediction. aux_classifier (nn.Module, optional): auxiliary classifier used during training N)__name__ __module__ __qualname____doc__ZC:\Projects\Historical\venv\Lib\site-packages\torchvision/models/segmentation/deeplabv3.pyrrs   Dr+rc >eZdZddededeeddffd ZxZS) DeepLabHead $ in_channels num_classes atrous_ratesreturnNc  tt||tjdddddtjdtjtjd|ddS)Nr rF)paddingbias)super__init__ASPPrConv2d BatchNorm2dReLU)selfr3r4r5 __class__s r,r<zDeepLabHead.__init__1su  l + + Ic315 9 9 9 N3   GII Ic; * *      r+)r/)r&r'r(intrr< __classcell__rBs@r,r.r.0sd  C c RU jn          r+r.c0eZdZdedededdffd ZxZS)ASPPConvr3 out_channelsdilationr6Nctj||d||dtj|tjg}t j|dS)Nr F)r9rIr:)rr>r?r@r;r<)rAr3rHrImodulesrBs r,r<zASPPConv.__init__<sW Ik<Hx^c d d d N< ( ( GII  '""""r+)r&r'r(rCr<rDrEs@r,rGrG;sY#C#s#c#d##########r+rGcPeZdZdededdffd ZdejdejfdZxZS) ASPPPoolingr3rHr6Nc ttjdtj||ddtj|tjdS)NrFr:)r;r<rAdaptiveAvgPool2dr>r?r@)rAr3rHrBs r,r<zASPPPooling.__init__Fsc    # # Ik< ? ? ? N< ( ( GII      r+xcp|jdd}|D] }||}tj||ddS)NbilinearF)sizemode align_corners)shapeF interpolate)rArQrUmods r,forwardzASPPPooling.forwardNsHwrss|  CAAA}QT %PPPPr+) r&r'r(rCr<torchTensorr\rDrEs@r,rMrMEs C s t      QQ%,QQQQQQQQr+rMc beZdZd dedeededdffd Zdejdejfd ZxZ S) r=r8r3r5rHr6Nc tg}|tjtj||ddtj|tjt|}|D]&}|t|||'|t||tj ||_ tjtjt|j |z|ddtj|tjtjd|_dS)NrFrOg?)r;r<appendr Sequentialr>r?r@tuplerGrM ModuleListconvslenDropoutproject)rAr3r5rHrKratesraterBs r,r<z ASPP.__init__Vs6  M")KquMMMr~^jOkOkmomtmvmv w w   l## F FD NN8KtDD E E E E{; ==>>>]7++ } Ic$*oo 4lAE R R R N< ( ( GII JsOO    r+rQcg}|jD] }|||!tj|d}||S)Nr)dim)rerar]catrh)rArQ_resconvress r,r\z ASPP.forwardls[J ! !D KKQ i!$$$||C   r+)r8) r&r'r(rCrr<r]r^r\rDrEs@r,r=r=Us  C x} TW bf      ,!!%,!!!!!!!!r+r=backboner4auxr6cddi}|rd|d<t||}|rtd|nd}td|}t|||S)Nlayer4outrrlayer3 return_layersii)rrr.r)rqr4rrrxaux_classifier classifiers r,_deeplabv3_resnetr{tsl u%M ("' h&x}MMMH36@WT;///DNT;//J Xz> : ::r+)rrz These weights were trained on a subset of COCO, using only the 20 categories that are present in the Pascal VOC dataset. ) categoriesmin_size_docsc beZdZedeediedddddd id d d  ZeZdS)rzHhttps://download.pytorch.org/models/deeplabv3_resnet50_coco-cd0a2569.pth resize_sizeijzVhttps://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_resnet50COCO-val2017-VOC-labelsgP@皙W@miou pixel_accgvWf@gGzd@ num_paramsrecipe_metrics_ops _file_sizeurl transformsmetaN r&r'r(r rr _COMMON_METACOCO_WITH_VOC_LABELS_V1DEFAULTr*r+r,rrs%g V7/SAAA    "n) !%,, !    "&GGGr+rc beZdZedeediedddddd id d d  ZeZdS)r zIhttps://download.pytorch.org/models/deeplabv3_resnet101_coco-586e9e4e.pthrrijzQhttps://github.com/pytorch/vision/tree/main/references/segmentation#fcn_resnet101rgP@rrgS+p@gm&m@rrNrr*r+r,r r s%g W7/SAAA    "i) !%,, !    "&GGGr+r c beZdZedeediedddddd id d d  ZeZdS)r!zMhttps://download.pytorch.org/models/deeplabv3_mobilenet_v3_large-fc3c493d.pthrriPKz`https://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_mobilenet_v3_largergfffff&N@gV@rgCl$@gJ +&E@rrNrr*r+r,r!r!s%g [7/SAAA    "x) !%,,      "&GGGr+r!c|j}dgdt|Dzt|dz gz}|d}||j}|d}||j}t |di}|rd|t |<t ||}|rt ||nd} t||} t|| | S) Nrc:g|]\}}t|dd|S)_is_cnF)getattr).0ibs r, z*_deeplabv3_mobilenetv3..s.\\\A8UZ@[@[\1\\\r+rrurrrw) features enumeraterfrHstrrrr.r) rqr4rr stage_indicesout_pos out_inplanesaux_pos aux_inplanesrxryrzs r,_deeplabv3_mobilenetv3rs  HC\\8)<)<\\\\`cdl`m`mpq`q_rrMBGG$1LBGG$1L\\5)M ,&+ c'll#&x}MMMH;>HW\;777DN\;77J Xz> : ::r+ pretrainedpretrained_backbone)weightsweights_backboneT)rprogressr4aux_lossrrrrrkwargsc t|}tj|}|=d}td|t |jd}td|d}n|d}t |gd}t|||}|*|| |d |S) adConstructs a DeepLabV3 model with a ResNet-50 backbone. .. betastatus:: segmentation module Reference: `Rethinking Atrous Convolution for Semantic Image Segmentation `__. Args: weights (:class:`~torchvision.models.segmentation.DeepLabV3_ResNet50_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.segmentation.DeepLabV3_ResNet50_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. num_classes (int, optional): number of output classes of the model (including the background) aux_loss (bool, optional): If True, it uses an auxiliary loss weights_backbone (:class:`~torchvision.models.ResNet50_Weights`, optional): The pretrained weights for the backbone **kwargs: unused .. autoclass:: torchvision.models.segmentation.DeepLabV3_ResNet50_Weights :members: Nr4r|rTFTTrreplace_stride_with_dilationr check_hash) rverifyrrrfrrr{load_state_dictget_state_dictrrr4rrrrqmodels r,r#r#sJ)//88G'./?@@+M;GLYeLfHgHghh (XtDD    0ObObObcccH h X > >E g44hSW4XXYYY Lr+c t|}tj|}|=d}td|t |jd}td|d}n|d}t |gd}t|||}|*|| |d |S) aiConstructs a DeepLabV3 model with a ResNet-101 backbone. .. betastatus:: segmentation module Reference: `Rethinking Atrous Convolution for Semantic Image Segmentation `__. Args: weights (:class:`~torchvision.models.segmentation.DeepLabV3_ResNet101_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.segmentation.DeepLabV3_ResNet101_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. num_classes (int, optional): number of output classes of the model (including the background) aux_loss (bool, optional): If True, it uses an auxiliary loss weights_backbone (:class:`~torchvision.models.ResNet101_Weights`, optional): The pretrained weights for the backbone **kwargs: unused .. autoclass:: torchvision.models.segmentation.DeepLabV3_ResNet101_Weights :members: Nr4r|rTrrrr) r rrrrfrrr{rrrs r,r$r$sJ*0099G(/0@AA+M;GLYeLfHgHghh (XtDD   !1PcPcPcdddH h X > >E g44hSW4XXYYY Lr+c t|}tj|}|=d}td|t |jd}td|d}n|d}t |d}t|||}|*|| |d|S) akConstructs a DeepLabV3 model with a MobileNetV3-Large backbone. Reference: `Rethinking Atrous Convolution for Semantic Image Segmentation `__. Args: weights (:class:`~torchvision.models.segmentation.DeepLabV3_MobileNet_V3_Large_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.segmentation.DeepLabV3_MobileNet_V3_Large_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. num_classes (int, optional): number of output classes of the model (including the background) aux_loss (bool, optional): If True, it uses an auxiliary loss weights_backbone (:class:`~torchvision.models.MobileNet_V3_Large_Weights`, optional): The pretrained weights for the backbone **kwargs: unused .. autoclass:: torchvision.models.segmentation.DeepLabV3_MobileNet_V3_Large_Weights :members: Nr4r|rTr)rdilatedr) r!rrrrfrrrrrrs r,r"r"SsF399'BBG189IJJ+M;GLYeLfHgHghh (XtDD   !*:DIIIH "8[( C CE g44hSW4XXYYY Lr+)9 functoolsrtypingrrrr]rtorch.nnrrYtransforms._presetsr _apir r r_metar_utilsrrr mobilenetv3rrrresnetrrrrrrfcnr__all__rrbr.rGrMModuler=rCboolr{rrr r!rr IMAGENET1K_V1r#r$r"r*r+r,rs********** $$$$$$7777777777777777######\\\\\\\\\\UUUUUUUUUUUUUUUUUUUUUUUU,,,,,,        (   &     "-   #####r}### Q Q Q Q Q"- Q Q Q !!!!!29!!!> ; ; ; $ ; ; ; ; ; " &&&&&&&&*&&&&&+&&&*&&&&&;&&&*;;; $; ;;;;. 5M N+-=-KL 59!%#3C3Q 000 0 100# 0 tn 0 /0 00000  0f 6N O+->-LM 6:!%#4E4S 000 1 200# 0 tn 0 01 00000  0f ?W X+-G-UV ?C!%#=W=e ... : ;..# . tn . 9: .....  ...r+