a d@: @sddlmZddlmZmZmZddlZddlmZddlm Z ddl m Z dd l mZmZmZdd lmZdd lmZmZmZdd lmZmZmZdd lmZmZmZmZm Z ddlm!Z!ddl"m#Z#gdZ$Gddde!Z%Gdddej&Z'Gdddej&Z(Gdddej&Z)Gdddej*Z+ee,ee-e%dddZ.edd d!Z/Gd"d#d#eZ0Gd$d%d%eZ1Gd&d'd'eZ2ee,ee-e%dd(d)Z3eed*e0j4fd+e j5fd,dd-dde j5d.ee0e-ee,ee-ee ee%d/d0d1Z6eed*e1j4fd+ej5fd,dd-ddej5d.ee1e-ee,ee-eeee%d/d2d3Z7eed*e2j4fd+ej5fd,dd-ddej5d.ee2e-ee,ee-eeee%d/d4d5Z8dS)6)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(r/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/torchvision/models/segmentation/deeplabv3.pyrsrcs&eZdZeeddfdd ZZS) DeepLabHeadN) in_channels num_classesreturnc 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*0sr*cs(eZdZeeeddfdd ZZS)ASPPConvN)r+ out_channelsdilationr-cs6tj||d||ddt|tg}tj|dS)NrF)r2rBr3)rr7r8r9r4r5)r:r+rArBmodulesr;r(r)r5<s zASPPConv.__init__r=r(r(r;r)r@;sr@cs:eZdZeeddfdd ZejejdddZZS) ASPPPoolingN)r+rAr-c s4ttdtj||dddt|tdS)NrFr3)r4r5rZAdaptiveAvgPool2dr7r8r9)r:r+rAr;r(r)r5Fs zASPPPooling.__init__xr-cCs2|jdd}|D] }||}qtj||dddS)NZbilinearF)sizemodeZ align_corners)shapeFZ interpolate)r:rGrImodr(r(r)forwardNs zASPPPooling.forward) r$r%r&r>r5torchTensorrNr?r(r(r;r)rDEsrDcsBeZdZd eeeeddfdd ZejejdddZZ S) r6r1N)r+ atrous_ratesrAr-c stg}|ttj||dddt|tt|}|D]}|t |||qF|t ||t ||_ ttjt |j ||dddt|ttd|_dS)NrFrEg?)r4r5appendr Sequentialr7r8r9tupler@rDZ ModuleListconvslenZDropoutproject)r:r+rQrArCZratesZrater;r(r)r5Vs  $ z ASPP.__init__rFcCs6g}|jD]}|||q tj|dd}||S)Nr)Zdim)rUrRrOcatrW)r:rGZ_resconvresr(r(r)rNls  z ASPP.forward)r1) r$r%r&r>rr5rOrPrNr?r(r(r;r)r6Usr6)backboner,auxr-cCsHddi}|rd|d<t||d}|r.td|nd}td|}t|||S)NZlayer4outr\Zlayer3 return_layersii)rrr*r)r[r,r\r_aux_classifier classifierr(r(r)_deeplabv3_resnetts  rb)rrz These weights were trained on a subset of COCO, using only the 20 categories that are present in the Pascal VOC dataset. ) categoriesZmin_sizeZ_docsc @sDeZdZedeeddiedddddd id d d d ZeZdS)rzHhttps://download.pytorch.org/models/deeplabv3_resnet50_coco-cd0a2569.pthZ resize_sizeijzVhttps://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_resnet50COCO-val2017-VOC-labelsgP@皙W@ZmiouZ pixel_accgvWf@gGzd@Z num_paramsZrecipeZ_metricsZ_ops _file_sizeurlZ transformsmetaN r$r%r&r rr _COMMON_METACOCO_WITH_VOC_LABELS_V1DEFAULTr(r(r(r)rs$ rc @sDeZdZedeeddiedddddd id d d d ZeZdS)rzIhttps://download.pytorch.org/models/deeplabv3_resnet101_coco-586e9e4e.pthrdreijzQhttps://github.com/pytorch/vision/tree/main/references/segmentation#fcn_resnet101rfgP@rgrhgS+p@gm&m@rirkNrnr(r(r(r)rs$ rc @sDeZdZedeeddiedddddd id d d d ZeZdS)r zMhttps://download.pytorch.org/models/deeplabv3_mobilenet_v3_large-fc3c493d.pthrdreiPKz`https://github.com/pytorch/vision/tree/main/references/segmentation#deeplabv3_mobilenet_v3_largerfgfffff&N@gV@rhgCl$@gJ +&E@rirkNrnr(r(r(r)r s$ r c Cs|j}dgddt|Dt|dg}|d}||j}|d}||j}t|di}|rld|t|<t||d }|rt||nd} t||} t|| | S) NrcSs g|]\}}t|ddr|qS)Z_is_cnF)getattr).0ibr(r(r) z*_deeplabv3_mobilenetv3..rr]r\r^) features enumeraterVrAstrrrr*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)r~progressr,aux_lossr)r~rr,rrkwargsr-cKst|}t|}|durDd}td|t|jd}td|d}n |durPd}t|gdd}t|||}|dur||j |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: Nr,rcrTFTTr~Zreplace_stride_with_dilationr) rverifyrrrVrmrrbload_state_dictget_state_dictr~rr,rrrr[modelr(r(r)r"s%   r"cKst|}t|}|durDd}td|t|jd}td|d}n |durPd}t|gdd}t|||}|dur||j |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: Nr,rcrTrrrr) rrrrrVrmrrbrrrr(r(r)r#s%   r#cKst|}t|}|durDd}td|t|jd}td|d}n |durPd}t|dd}t|||}|dur||j |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: Nr,rcrTr)r~Zdilatedr) r rrrrVrmrr}rrrr(r(r)r!Ss#    r!)9 functoolsrtypingrrrrOrZtorch.nnrrLZtransforms._presetsr Z_apir r r Z_metar_utilsrrrZ mobilenetv3rrrZresnetrrrrrrZfcnr__all__rrSr*r@rDModuler6r>boolrbrorrr r}rpZ IMAGENET1K_V1r"r#r!r(r(r(r)s              33