o ϯi#@sddlmZmZmZddlZddlmZddlmZddl m Z Gdddej Z dddd d dd d ej d d ejd dededededeedeedeededeededededededej fddZGd d!d!ej ZdS)")CallableOptionalTupleN)set_attributes) _size_3_tcseZdZdZdddddddddededeedeeed eeed eeed eeed eeed eee deddffddZ de j de j fddZ ZS) ConvReduce3Du Builds a list of convolutional operators and performs summation on the outputs. :: Conv3d, Conv3d, ..., Conv3d ↓ Sum Nsum)stridepadding padding_modedilationgroupsbiasreduction_method in_channels out_channels kernel_sizer r r r r rrreturnc s(t| dvs J| |_g} tt|D]u} |||| d} |dur0|| dur0|| | d<|dur@|| dur@|| | d<|durP|| durP|| | d<|dur`|| dur`|| | d<| durp| | durp| | | d<|dur|| dur|| | d <| tjd i| qt| |_ dS) a Args: in_channels int: number of input channels. out_channels int: number of output channels produced by the convolution(s). kernel_size tuple(_size_3_t): Tuple of sizes of the convolutionaling kernels. stride tuple(_size_3_t): Tuple of strides of the convolutions. padding tuple(_size_3_t): Tuple of paddings added to all three sides of the input. padding_mode tuple(string): Tuple of padding modes for each convs. Options include `zeros`, `reflect`, `replicate` or `circular`. dilation tuple(_size_3_t): Tuple of spacings between kernel elements. groups tuple(_size_3_t): Tuple of numbers of blocked connections from input channels to output channels. bias tuple(bool): If `True`, adds a learnable bias to the output. reduction_method str: Options include `sum` and `cat`. )rcat)rrrNr r r r rr ) super__init__rrangelenappendnnConv3d ModuleListconvs)selfrrrr r r r r rr conv_listind conv_param __class__rT/home/ubuntu/.local/lib/python3.10/site-packages/pytorchvideo/layers/convolutions.pyrs.        zConvReduce3D.__init__xcCsng}tt|jD] }||j||q |jdkr)tj|ddjddd}|S|jdkr5tj|dd}|S)Nrr)dimF)r'keepdimr) rrrrrtorchstackrr)rr&outputr!rrr%forwardMs  zConvReduce3D.forward)__name__ __module__ __qualname____doc__intrrrstrboolrr*Tensorr- __classcell__rrr#r%r sB        7rF)r7r7)r8r8)r)r)r)r)gh㈵>g?) inner_channels conv_xy_firstrr r rr r normnorm_eps norm_momentum activationrrr9r:rr r rr r r;r<r=r>rc Cs|dur|}| dksJdt|dkrt|dksJdtj|s$|n||s)|n||dddf|dddf|dddf|d}| durFdn| || | d}| durSdn| }tj|s\|n||sa|n|d|d|dfd|d|dfd|d|df|d}t|||||d S) uB Create a 2plus1d conv layer. It performs spatiotemporal Convolution, BN, and Relu following by a spatiotemporal pooling. :: Conv_t (or Conv_xy if conv_xy_first = True) ↓ Normalization ↓ Activation ↓ Conv_xy (or Conv_t if conv_xy_first = True) Normalization options include: BatchNorm3d and None (no normalization). Activation options include: ReLU, Softmax, Sigmoid, and None (no activation). Args: in_channels (int): input channel size of the convolution. out_channels (int): output channel size of the convolution. kernel_size (tuple): convolutional kernel size(s). stride (tuple): convolutional stride size(s). padding (tuple): convolutional padding size(s). bias (bool): convolutional bias. If true, adds a learnable bias to the output. groups (int): Number of groups in convolution layers. value >1 is unsupported. dilation (tuple): dilation value in convolution layers. value >1 is unsupported. conv_xy_first (bool): If True, spatial convolution comes before temporal conv norm (callable): a callable that constructs normalization layer, options include nn.BatchNorm3d, None (not performing normalization). norm_eps (float): normalization epsilon. norm_momentum (float): normalization momentum. activation (callable): a callable that constructs activation layer, options include: nn.ReLU, nn.Softmax, nn.Sigmoid, and None (not performing activation). Returns: (nn.Module): 2plus1d conv layer. Nr)z?Support for groups is not implemented in R2+1 convolution layerzBSupport for dillaiton is not implemented in R2+1 convolution layerr)rrrr r r) num_featuresepsmomentumr8conv_tr;r>conv_xyr:)maxminrr Conv2plus1d)rrr9r:rr r rr r r;r<r=r> conv_t_module norm_moduleactivation_moduleconv_xy_modulerrr%create_conv_2plus1dXsH=           rLc sfeZdZdZdddddddejdejdejdejd ed df fd d Zd ej d ej fddZ Z S)rGu Implementation of 2+1d Convolution by factorizing 3D Convolution into an 1D temporal Convolution and a 2D spatial Convolution with Normalization and Activation module in between: :: Conv_t (or Conv_xy if conv_xy_first = True) ↓ Normalization ↓ Activation ↓ Conv_xy (or Conv_t if conv_xy_first = True) The 2+1d Convolution is used to build the R(2+1)D network. NFrBrCr;r>rDr:rcs6tt|t|jdusJ|jdusJdS)ar Args: conv_t (torch.nn.modules): temporal convolution module. norm (torch.nn.modules): normalization module. activation (torch.nn.modules): activation module. conv_xy (torch.nn.modules): spatial convolution module. conv_xy_first (bool): If True, spatial convolution comes before temporal conv N)rrrlocalsrCrD)rrCr;r>rDr:r#rr%rs  zConv2plus1d.__init__r&cCsd|jr||n||}|jr||n|}|jr||n|}|jr+||}|S||}|S)N)r:rDrCr;r>)rr&rrr%r-s zConv2plus1d.forward) r.r/r0r1rModuler4rr*r5r-r6rrr#r%rGs*rG)typingrrrr*torch.nnrpytorchvideo.layers.utilsrtorch.nn.common_typesrrNr BatchNorm3dReLUr2r4floatrLrGrrrr%sd   R     g