U (d( @sddlZddlmZmZmZmZmZddlmZm Z ddl m Z ddl mZmZmZddlmZmZddlmZmZdd lmZmZGd d d ejZed d dfde jddddeeeedejfeeeeeeedddZd#ej eeeeeeeedejfedddZ!e"eeeeedddZ#ed ddfde jddddeeee"edejfeeeeeeejdddZ$d$eej%ej&fe"eeeeeeeedejfejd d!d"Z'dS)%N)CallableDictListOptionalUnion)nnTensor)misc) ExtraFPNBlockFeaturePyramidNetworkLastLevelMaxPool) mobilenetresnet) WeightsEnum_get_enum_from_fn)IntermediateLayerGetterhandle_legacy_interfacec sleZdZdZd ejeeefee e e e e e dejfddfdd Z eeeefddd ZZS) BackboneWithFPNa Adds a FPN on top of a model. Internally, it uses torchvision.models._utils.IntermediateLayerGetter to extract a submodel that returns the feature maps specified in return_layers. The same limitations of IntermediateLayerGetter apply here. Args: backbone (nn.Module) return_layers (Dict[name, new_name]): a dict containing the names of the modules for which the activations will be returned as the key of the dict, and the value of the dict is the name of the returned activation (which the user can specify). in_channels_list (List[int]): number of channels for each feature map that is returned, in the order they are present in the OrderedDict out_channels (int): number of channels in the FPN. norm_layer (callable, optional): Module specifying the normalization layer to use. Default: None Attributes: out_channels (int): the number of channels in the FPN N.)backbone return_layersin_channels_list out_channels extra_blocks norm_layerreturncsBt|dkrt}t||d|_t||||d|_||_dS)N)r)rrrr)super__init__r rbodyr fpnr)selfrrrrrr __class__O/tmp/pip-unpacked-wheel-vx7f76es/torchvision/models/detection/backbone_utils.pyr!s zBackboneWithFPN.__init__)xrcCs||}||}|S)N)rr)r r%r#r#r$forward8s  zBackboneWithFPN.forward)NN)__name__ __module__ __qualname____doc__rModulerstrrintrr rrrr& __classcell__r#r#r!r$r s rZ pretrainedcCsttj|ddSN backbone_nameZ IMAGENET1K_V1)rr__dict__from_strkwargsr#r#r$Ar5)weights)rtrainable_layersreturned_layersr.)r0r7rr9r:rrcCs tj|||d}t||||S)a Constructs a specified ResNet backbone with FPN on top. Freezes the specified number of layers in the backbone. Examples:: >>> from torchvision.models.detection.backbone_utils import resnet_fpn_backbone >>> backbone = resnet_fpn_backbone('resnet50', weights=ResNet50_Weights.DEFAULT, trainable_layers=3) >>> # get some dummy image >>> x = torch.rand(1,3,64,64) >>> # compute the output >>> output = backbone(x) >>> print([(k, v.shape) for k, v in output.items()]) >>> # returns >>> [('0', torch.Size([1, 256, 16, 16])), >>> ('1', torch.Size([1, 256, 8, 8])), >>> ('2', torch.Size([1, 256, 4, 4])), >>> ('3', torch.Size([1, 256, 2, 2])), >>> ('pool', torch.Size([1, 256, 1, 1]))] Args: backbone_name (string): resnet architecture. Possible values are 'resnet18', 'resnet34', 'resnet50', 'resnet101', 'resnet152', 'resnext50_32x4d', 'resnext101_32x8d', 'wide_resnet50_2', 'wide_resnet101_2' weights (WeightsEnum, optional): The pretrained weights for the model norm_layer (callable): it is recommended to use the default value. For details visit: (https://github.com/facebookresearch/maskrcnn-benchmark/issues/267) trainable_layers (int): number of trainable (not frozen) layers starting from final block. Valid values are between 0 and 5, with 5 meaning all backbone layers are trainable. returned_layers (list of int): The layers of the network to return. Each entry must be in ``[1, 4]``. By default all layers are returned. extra_blocks (ExtraFPNBlock or None): if provided, extra operations will be performed. It is expected to take the fpn features, the original features and the names of the original features as input, and returns a new list of feature maps and their corresponding names. By default a ``LastLevelMaxPool`` is used. r7r)rr1_resnet_fpn_extractor)r0r7rr9r:rrr#r#r$resnet_fpn_backbone>s2r=)rr9r:rrrc s|dks|dkrtd|dddddgd|}|dkrF|d |D](\}tfd d |DrN|d qN|dkrt}|dkrd dddg}t|dkst|dkrtd|ddt|D}|j dfdd |D}d} t |||| ||dS)Nrz3Trainable layers should be in the range [0,5], got Zlayer4Zlayer3Zlayer2Zlayer1Zconv1Zbn1csg|]}| qSr#) startswith).0layer)namer#r$ sz)_resnet_fpn_extractor..Fr r8z6Each returned layer should be in the range [1,4]. Got cSs i|]\}}d|t|qS)rAr,r@vkr#r#r$ sz)_resnet_fpn_extractor..csg|]}d|dqS)r rDr#r@i)in_channels_stage2r#r$rCsrr) ValueErrorappendZnamed_parametersallrequires_grad_r minmax enumerateZinplanesr) rr9r:rrZlayers_to_train parameterrrrr#)rNrBr$r<ts4     r<) is_trainedtrainable_backbone_layers max_value default_valuercCsX|s"|dk rtd|d|}|dkr.|}|dks>||krTtd|d|d|S)NzChanging trainable_backbone_layers has not effect if neither pretrained nor pretrained_backbone have been set to True, falling back to trainable_backbone_layers=z! so that all layers are trainablerz4Trainable backbone layers should be in the range [0,], got  )warningswarnrQ)rYrZr[r\r#r#r$_validate_trainable_layerss racCsttj|ddSr/)rrr1r2r3r#r#r$r5r6)r0r7rrr9r:rrcCs"tj|||d}t|||||S)Nr;)rr1_mobilenet_extractor)r0r7rrr9r:rrr#r#r$mobilenet_backbonesrc)rrr9r:rrrcsljdgddtDtdgt}|dksD||krZtd|d|d|dkrjtn ||}d|D]}|D]} | dqqd } |rB|dkrt}|dkr|d |dg}t|dkst||krtd |dd|dfd d t|D} fdd|D} t | | | ||dSt t dj | d} | | _ | SdS)NrcSs g|]\}}t|ddr|qS)Z_is_cnF)getattr)r@rMbr#r#r$rCs z(_mobilenet_extractor..rDz+Trainable layers should be in the range [0,r]r^FrOr z.Each returned layer should be in the range [0,cs i|]\}}|t|qSr#rFrG) stage_indicesr#r$rJsz(_mobilenet_extractor..csg|]}|jqSr#)rrLrrfr#r$rCsrP)featuresrWlenrQ parametersrTr rUrVrrZ SequentialZConv2dr)rrr9r:rrZ num_stagesZ freeze_beforererXrrrmr#rgr$rbsB& rb)NNN)NNN)(r_typingrrrrrZtorchrrZtorchvision.opsr Z misc_nn_opsZ'torchvision.ops.feature_pyramid_networkr r r rrZ_apirr_utilsrrr+rZFrozenBatchNorm2dr,r-r=ZResNetr<boolrarcZ MobileNetV2Z MobileNetV3rbr#r#r#r$s 1   3  $