o )is:@sddlmZddlmZmZmZddlZddlmZddlm Z ddl m Z dd l mZmZmZdd lmZdd lmZmZmZdd lmZmZmZdd lmZmZmZmZm Z ddlm!Z!ddl"m#Z#gdZ$Gddde!Z%Gdddej&Z'Gdddej&Z(Gdddej&Z)Gdddej*Z+dede,dee-de%fd d!Z.ed"d#d$Z/Gd%d&d&eZ0Gd'd(d(eZ1Gd)d*d*eZ2dede,dee-de%fd+d,Z3eed-e0j4fd.e j5fd/dd0dde j5d1d2ee0d3e-dee,d4ee-d5ee d6ede%fd7d8Z6eed-e1j4fd.ej5fd/dd0ddej5d1d2ee1d3e-dee,d4ee-d5eed6ede%fd9d:Z7eed-e2j4fd.ej5fd/dd0ddej5d1d2ee2d3e-dee,d4ee-d5eed6ede%fd;d<Z8dS)=)partial)AnyListOptionalN)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_resnet101c@seZdZdZdS)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__r(r(x/var/www/html/eduruby.in/lip-sync/lip-sync-env/lib/python3.10/site-packages/torchvision/models/segmentation/deeplabv3.pyrsrcs*eZdZdededdffdd ZZS) DeepLabHead in_channels num_classesreturnNc sFtt|gdtjddddddtdttd|ddS)N) $rrF)paddingbias)super__init__ASPPrConv2d BatchNorm2dReLU)selfr+r, __class__r(r)r51s  zDeepLabHead.__init__r$r%r&intr5 __classcell__r(r(r;r)r*0s"r*cs.eZdZdedededdffdd ZZS)ASPPConvr+ out_channelsdilationr-Ncs6tj||d||ddt|tg}tj|dS)NrF)r2rBr3)rr7r8r9r4r5)r:r+rArBmodulesr;r(r)r5<s zASPPConv.__init__r=r(r(r;r)r@;s&r@cs@eZdZdededdffdd Zdejdejfdd ZZS) ASPPPoolingr+rAr-Nc s4ttdtj||dddt|tdS)NrFr3)r4r5rZAdaptiveAvgPool2dr7r8r9)r:r+rAr;r(r)r5Fs zASPPPooling.__init__xcCs2|jdd}|D]}||}q tj||dddS)NZbilinearF)sizemodeZ align_corners)shapeFZ interpolate)r:rFrHmodr(r(r)forwardNs zASPPPooling.forward) r$r%r&r>r5torchTensorrMr?r(r(r;r)rDEsrDc sJeZdZd dedeededdffdd Zd ejdejfd d ZZ S) r6r1r+ atrous_ratesrAr-Nc stg}|ttj||dddt|tt|}|D] }|t |||q#|t ||t ||_ ttjt |j ||dddt|ttd|_dS)NrFrEg?)r4r5appendr Sequentialr7r8r9tupler@rDZ ModuleListconvslenZDropoutproject)r:r+rPrArCZratesZrater;r(r)r5Vs $  z ASPP.__init__rFcCs6g}|jD] }|||qtj|dd}||S)Nr)dim)rTrQrNcatrV)r:rFZ_resconvresr(r(r)rMls   z ASPP.forward)r1) r$r%r&r>rr5rNrOrMr?r(r(r;r)r6Us$r6backboner,auxr-cCsHddi}|r d|d<t||d}|rtd|nd}td|}t|||S)NZlayer4outr\Zlayer3 return_layersii)rrr*r)r[r,r\r_aux_classifier classifierr(r(r)_deeplabv3_resnetts   rb)rrz These weights were trained on a subset of COCO, using only the 20 categories that are present in the Pascal VOC dataset. ) categoriesZmin_sizeZ_docsc @DeZdZedeeddiedddddd id d d d ZeZdS)rzHhttps://download.pytorch.org/models/deeplabv3_resnet50_coco-cd0a2569.pthZ resize_sizeijzVhttps://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_resnet50COCO-val2017-VOC-labelsgP@皙W@ZmiouZ pixel_accgvWf@gGzd@Z num_paramsZrecipeZ_metricsZ_ops _file_sizeurlZ transformsmetaN r$r%r&r rr _COMMON_METACOCO_WITH_VOC_LABELS_V1DEFAULTr(r(r(r)r& rc @rd)rzIhttps://download.pytorch.org/models/deeplabv3_resnet101_coco-586e9e4e.pthrerfijzQhttps://github.com/pytorch/vision/tree/main/references/segmentation#fcn_resnet101rggP@rhrigS+p@gm&m@rjrlNror(r(r(r)rrsrc @rd)r zMhttps://download.pytorch.org/models/deeplabv3_mobilenet_v3_large-fc3c493d.pthrerfiPKz`https://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_mobilenet_v3_largerggfffff&N@gV@rigCl$@gJ +&E@rjrlNror(r(r(r)r rsr c Cs|j}dgddt|Dt|dg}|d}||j}|d}||j}t|di}|r6d|t|<t||d }|rCt||nd} t||} t|| | S) NrcSs g|] \}}t|ddr|qS)Z_is_cnF)getattr).0ibr(r(r) s z*_deeplabv3_mobilenetv3..rr]r\r^) features enumeraterUrAstrrrr*r) r[r,r\Z stage_indicesZout_posZ out_inplanesZaux_posZ aux_inplanesr_r`rar(r(r)_deeplabv3_mobilenetv3s&       r~Z pretrainedZpretrained_backbone)weightsweights_backboneT)rprogressr,aux_lossrrrrrkwargscKt|}t|}|dur"d}td|t|jd}td|d}n|dur(d}t|gdd}t|||}|durD||j |dd |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: Nr,rcrTFTTrZreplace_stride_with_dilationrZ check_hash) rverifyrrrUrnrrbload_state_dictget_state_dictrrr,rrrr[modelr(r(r)r" %  r"cKr) 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: Nr,rcrTrrrr) rrrrrUrnrrbrrrr(r(r)r#rr#cKst|}t|}|dur"d}td|t|jd}td|d}n|dur(d}t|dd}t|||}|durB||j |dd|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: Nr,rcrTr)rZdilatedr) r rrrrUrnrr~rrrr(r(r)r!Ss #   r!)9 functoolsrtypingrrrrNrZtorch.nnrrKZtransforms._presetsr Z_apir r r _metar_utilsrrrZ mobilenetv3rrrZresnetrrrrrrZfcnr__all__rrRr*r@rDModuler6r>boolrbrprrr r~rqZ IMAGENET1K_V1r"r#r!r(r(r(r)s             33