o Qe%@sPddlmZddlmZgZGdddeZGdddeZGdddeZd S) )Layer) functionalc2eZdZdZd fdd ZddZdd ZZS) PixelShufflea PixelShuffle Layer Rearranges elements in a tensor of shape :math:`[N, C, H, W]` to a tensor of shape :math:`[N, C/upscale_factor^2, H*upscale_factor, W imes upscale_factor]`, or from shape :math:`[N, H, W, C]` to :math:`[N, H imes upscale_factor, W imes upscale_factor, C/upscale_factor^2]`. This is useful for implementing efficient sub-pixel convolution with a stride of 1/upscale_factor. Please refer to the paper: `Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network `_ . by Shi et. al (2016) for more details. Parameters: upscale_factor(int): factor to increase spatial resolution. data_format (str, optional): The data format of the input and output data. An optional string from: "NCHW", "NHWC". The default is "NCHW". When it is "NCHW", the data is stored in the order of: [batch_size, input_channels, input_height, input_width]. name (str, optional): Name for the operation (optional, default is None). For more information, please refer to :ref:`api_guide_Name`. Shape: - x: 4-D tensor with shape of :math:`(N, C, H, W)` or :math:`(N, H, W, C)`. - out: 4-D tensor with shape of :math:`(N, C/upscale_factor^2, H imes upscale_factor, W imes upscale_factor)` or :math:`(N, H imes upscale_factor, W imes upscale_factor, C/upscale_factor^2)`. Examples: .. code-block:: python import paddle import paddle.nn as nn x = paddle.randn(shape=[2,9,4,4]) pixel_shuffle = nn.PixelShuffle(3) out_var = pixel_shuffle(x) out = out_var.numpy() print(out.shape) # (2, 1, 12, 12) NCHWNcsLtt|t|tstd|dvrtd|||_||_ ||_ dS)Nzupscale factor must be int typerZNHWCBData format should be 'NCHW' or 'NHWC'.But recevie data format: {}) superr__init__ isinstanceint TypeError ValueErrorformat_upscale_factor _data_format_name)selfZupscale_factor data_formatname __class__FD:\Projects\ConvertPro\env\Lib\site-packages\paddle/nn/layer/vision.pyr ?s  zPixelShuffle.__init__cCt||j|j|jSN)rZ pixel_shufflerrrrxrrrforwardM zPixelShuffle.forwardcCDd|j}|jdkr|d|j7}|jdur |d|j7}|S)Nzupscale_factor={}r, data_format={} , name={})rrrrrZmain_strrrr extra_reprQ   zPixelShuffle.extra_reprrN__name__ __module__ __qualname____doc__r rr$ __classcell__rrrrrs 'rcr) PixelUnshuffleaO Rearranges elements in a tensor of shape :math:`[N, C, H, W]` to a tensor of shape :math:`[N, r^2C, H/r, W/r]`, or from shape :math:`[N, H, W, C]` to :math:`[N, H/r, W/r, r^2C]`, where :math:`r` is the downscale factor. This operation is the reversion of PixelShuffle operation. Please refer to the paper: `Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network `_ . by Shi et. al (2016) for more details. Parameters: downscale_factor (int): Factor to decrease spatial resolution. data_format (str, optional): The data format of the input and output data. An optional string of NCHW or NHWC. The default is NCHW. When it is NCHW, the data is stored in the order of [batch_size, input_channels, input_height, input_width]. name (str, optional): Name for the operation (optional, default is None). Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Shape: - **x**: 4-D tensor with shape of :math:`[N, C, H, W]` or :math:`[N, C, H, W]`. - **out**: 4-D tensor with shape of :math:`[N, r^2C, H/r, W/r]` or :math:`[N, H/r, W/r, r^2C]`, where :math:`r` is :attr:`downscale_factor`. Examples: .. code-block:: python import paddle import paddle.nn as nn x = paddle.randn([2, 1, 12, 12]) pixel_unshuffle = nn.PixelUnshuffle(3) out = pixel_unshuffle(x) print(out.shape) # [2, 9, 4, 4] rNc\tt|t|tstd|dkrtd|dvr#td|||_||_ ||_ dS)Nz!Downscale factor must be int typez!Downscale factor must be positiverr) r r-r r r r rr_downscale_factorrr)rZdownscale_factorrrrrrr {  zPixelUnshuffle.__init__cCrr)rZpixel_unshuffler0rrrrrrrrzPixelUnshuffle.forwardcCr )Nzdownscale_factor={}rr!r")rr0rrr#rrrr$r%zPixelUnshuffle.extra_reprr&r'rrrrr-Zs  r-cr) ChannelShufflea This operator divides channels in a tensor of shape [N, C, H, W] or [N, H, W, C] into g groups, getting a tensor with the shape of [N, g, C/g, H, W] or [N, H, W, g, C/g], and transposes them as [N, C/g, g, H, W] or [N, H, W, g, C/g], then rearranges them to original tensor shape. This operation can improve the interaction between channels, using features efficiently. Please refer to the paper: `ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices `_ . by Zhang et. al (2017) for more details. Parameters: groups (int): Number of groups to divide channels in. data_format (str): The data format of the input and output data. An optional string of NCHW or NHWC. The default is NCHW. When it is NCHW, the data is stored in the order of [batch_size, input_channels, input_height, input_width]. name (str, optional): Name for the operation (optional, default is None). Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Shape: - **x**: 4-D tensor with shape of [N, C, H, W] or [N, H, W, C]. - **out**: 4-D tensor with shape and dtype same as x. Examples: .. code-block:: python import paddle import paddle.nn as nn x = paddle.arange(0, 0.6, 0.1, 'float32') x = paddle.reshape(x, [1, 6, 1, 1]) # [[[[0. ]], # [[0.10000000]], # [[0.20000000]], # [[0.30000001]], # [[0.40000001]], # [[0.50000000]]]] channel_shuffle = nn.ChannelShuffle(3) y = channel_shuffle(x) # [[[[0. ]], # [[0.20000000]], # [[0.40000001]], # [[0.10000000]], # [[0.30000001]], # [[0.50000000]]]] rNcr.)Nzgroups must be int typer/zgroups must be positiverr) r r2r r r r rr_groupsrr)rgroupsrrrrrr r1zChannelShuffle.__init__cCrr)rZchannel_shuffler3rrrrrrrszChannelShuffle.forwardcCr )Nz groups={}rr!r")rr3rrr#rrrr$r%zChannelShuffle.extra_reprr&r'rrrrr2s )r2N)rr__all__rr-r2rrrrs  C?