U (dw1@sddlZddlmZmZmZddlZddlmZddlmZej j j Z Gdddej j Z Gdd d ej jZGd d d eZGd d d eZGdddej j ZGdddej jZGdddej j ZdS)N)CallableListOptional)Tensor)_log_api_usage_oncecsreZdZdZdeedfdd Zeeee e ee ee edfdd Z e e d d d Z ed d dZZS)FrozenBatchNorm2da! BatchNorm2d where the batch statistics and the affine parameters are fixed Args: num_features (int): Number of features ``C`` from an expected input of size ``(N, C, H, W)`` eps (float): a value added to the denominator for numerical stability. Default: 1e-5 h㈵>) num_featuresepscsdtt|||_|dt||dt||dt||dt|dS)Nweightbias running_mean running_var)super__init__rr Zregister_buffertorchZoneszeros)selfr r  __class__8/tmp/pip-unpacked-wheel-vx7f76es/torchvision/ops/misc.pyrs zFrozenBatchNorm2d.__init__) state_dictprefixlocal_metadatastrict missing_keysunexpected_keys error_msgsc s2|d}||kr||=t|||||||dS)NZnum_batches_tracked)r_load_from_state_dict) rrrrrrrrZnum_batches_tracked_keyrrrr #s z'FrozenBatchNorm2d._load_from_state_dictxreturncCsr|jdddd}|jdddd}|jdddd}|jdddd}|||j}|||}|||S)N)r Zreshaper rrr Zrsqrt)rr"wbrvZrmscaler rrrforward5s zFrozenBatchNorm2d.forward)r#cCs$|jjd|jjdd|jdS)N(rz, eps=))r__name__r shaper )rrrr__repr__@szFrozenBatchNorm2d.__repr__)r )r- __module__ __qualname____doc__intfloatrdictstrboolrr rr*r/ __classcell__rrrrr s   rcseZdZddddejjejjdddejjf eeeee eee e dejj fe e dejj fee e e e e dejj fdd fdd Z ZS) ConvNormActivationr$NT.) in_channels out_channels kernel_sizestridepaddinggroups norm_layeractivation_layerdilationinplacer conv_layerr#c  s|dkr|dd| }| dkr(|dk} | |||||| || dg} |dk rX| |||dk r| dkrlind| i}| |f|tj| t|||_|jtkrtddS)Nr$r)rCr@r rDzhDon't use ConvNormActivation directly, please use Conv2dNormActivation and Conv3dNormActivation instead.) appendrrrr<rr9warningswarn)rr;r<r=r>r?r@rArBrCrDr rElayersparamsrrrrEs6   zConvNormActivation.__init__)r-r0r1rnn BatchNorm2dReLUConv2dr3rrModuler7rr8rrrrr9Ds2r9cseZdZdZddddejjejjdddf eeeee eee e dejj fe e dejj fee e e e dd fdd Z ZS) Conv2dNormActivationa Configurable block used for Convolution2d-Normalization-Activation blocks. Args: in_channels (int): Number of channels in the input image out_channels (int): Number of channels produced by the Convolution-Normalization-Activation block kernel_size: (int, optional): Size of the convolving kernel. Default: 3 stride (int, optional): Stride of the convolution. Default: 1 padding (int, tuple or str, optional): Padding added to all four sides of the input. Default: None, in which case it will calculated as ``padding = (kernel_size - 1) // 2 * dilation`` groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1 norm_layer (Callable[..., torch.nn.Module], optional): Norm layer that will be stacked on top of the convolution layer. If ``None`` this layer wont be used. Default: ``torch.nn.BatchNorm2d`` activation_layer (Callable[..., torch.nn.Module], optional): Activation function which will be stacked on top of the normalization layer (if not None), otherwise on top of the conv layer. If ``None`` this layer wont be used. Default: ``torch.nn.ReLU`` dilation (int): Spacing between kernel elements. Default: 1 inplace (bool): Parameter for the activation layer, which can optionally do the operation in-place. Default ``True`` bias (bool, optional): Whether to use bias in the convolution layer. By default, biases are included if ``norm_layer is None``. r:r$NT. r;r<r=r>r?r@rArBrCrDr r#c s*t||||||||| | | tjj dSN)rrrrKrN rr;r<r=r>r?r@rArBrCrDr rrrrszConv2dNormActivation.__init__)r-r0r1r2rrKrLrMr3rrrOr7rr8rrrrrPws0rPcseZdZdZddddejjejjdddf eeeee eee e dejj fe e dejj fee e e e dd fdd Z ZS) Conv3dNormActivationa Configurable block used for Convolution3d-Normalization-Activation blocks. Args: in_channels (int): Number of channels in the input video. out_channels (int): Number of channels produced by the Convolution-Normalization-Activation block kernel_size: (int, optional): Size of the convolving kernel. Default: 3 stride (int, optional): Stride of the convolution. Default: 1 padding (int, tuple or str, optional): Padding added to all four sides of the input. Default: None, in which case it will calculated as ``padding = (kernel_size - 1) // 2 * dilation`` groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1 norm_layer (Callable[..., torch.nn.Module], optional): Norm layer that will be stacked on top of the convolution layer. If ``None`` this layer wont be used. Default: ``torch.nn.BatchNorm3d`` activation_layer (Callable[..., torch.nn.Module], optional): Activation function which will be stacked on top of the normalization layer (if not None), otherwise on top of the conv layer. If ``None`` this layer wont be used. Default: ``torch.nn.ReLU`` dilation (int): Spacing between kernel elements. Default: 1 inplace (bool): Parameter for the activation layer, which can optionally do the operation in-place. Default ``True`` bias (bool, optional): Whether to use bias in the convolution layer. By default, biases are included if ``norm_layer is None``. r:r$NT.rQc s*t||||||||| | | tjj dSrR)rrrrKZConv3drSrrrrszConv3dNormActivation.__init__)r-r0r1r2rrKZ BatchNorm3drMr3rrrOr7rr8rrrrrTs0rTcsteZdZdZejjejjfeee dejj fe dejj fddfdd Z e e ddd Z e e dd d ZZS) SqueezeExcitationaE This block implements the Squeeze-and-Excitation block from https://arxiv.org/abs/1709.01507 (see Fig. 1). Parameters ``activation``, and ``scale_activation`` correspond to ``delta`` and ``sigma`` in eq. 3. Args: input_channels (int): Number of channels in the input image squeeze_channels (int): Number of squeeze channels activation (Callable[..., torch.nn.Module], optional): ``delta`` activation. Default: ``torch.nn.ReLU`` scale_activation (Callable[..., torch.nn.Module]): ``sigma`` activation. Default: ``torch.nn.Sigmoid`` .N)input_channelssqueeze_channels activationscale_activationr#csXtt|tjd|_tj||d|_tj||d|_ ||_ ||_ dS)Nr$) rrrrrKZAdaptiveAvgPool2davgpoolrNfc1fc2rXrY)rrVrWrXrYrrrrs zSqueezeExcitation.__init__)inputr#cCs2||}||}||}||}||SrR)rZr[rXr\rYrr]r)rrr_scales     zSqueezeExcitation._scalecCs||}||SrR)r_r^rrrr*s zSqueezeExcitation.forward)r-r0r1r2rrKrMZSigmoidr3rrOrrr_r*r8rrrrrUsrUc sjeZdZdZdejjdddfeeee e dejj fe e dejj fe e e e dfdd ZZS) MLPaThis block implements the multi-layer perceptron (MLP) module. Args: in_channels (int): Number of channels of the input hidden_channels (List[int]): List of the hidden channel dimensions norm_layer (Callable[..., torch.nn.Module], optional): Norm layer that will be stacked on top of the convolution layer. If ``None`` this layer wont be used. Default: ``None`` activation_layer (Callable[..., torch.nn.Module], optional): Activation function which will be stacked on top of the normalization layer (if not None), otherwise on top of the conv layer. If ``None`` this layer wont be used. Default: ``torch.nn.ReLU`` inplace (bool): Parameter for the activation layer, which can optionally do the operation in-place. Default ``True`` bias (bool): Whether to use bias in the linear layer. Default ``True`` dropout (float): The probability for the dropout layer. Default: 0.0 NTg.)r;hidden_channelsrArBrDr dropoutc s|dkr ind|i}g} |} |ddD]\} | tjj| | |d|dk rZ| || | |f|| tjj|f|| } q(| tjj| |d|d| tjj|f|tj| t|dS)NrDr%)r )rFrrKZLinearZDropoutrrr) rr;rarArBrDr rbrJrIZin_dimZ hidden_dimrrrrs  z MLP.__init__)r-r0r1r2rrKrMr3rrrrOr7r4rr8rrrrr`sr`cs:eZdZdZeedfdd ZeedddZZ S)PermutezThis module returns a view of the tensor input with its dimensions permuted. Args: dims (List[int]): The desired ordering of dimensions )dimscst||_dSrR)rrrd)rrdrrrr4s zPermute.__init__r!cCst||jSrR)rZpermuterd)rr"rrrr*8szPermute.forward) r-r0r1r2rr3rrr*r8rrrrrc-src)rGtypingrrrrrutilsrrKZ functionalZ interpolaterOrZ Sequentialr9rPrTrUr`rcrrrrs   7321',