# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. ### NOTE: the API of this file is based on Paddle1.8, the API in layers.py is based on Paddle2.0 import numpy as np import logging import paddle.fluid as fluid import paddle.fluid.core as core import paddle.fluid.dygraph_utils as dygraph_utils from paddle.fluid.data_feeder import check_variable_and_dtype from paddle.fluid.framework import _varbase_creator, in_dygraph_mode, _in_legacy_dygraph, _non_static_mode from paddle import _C_ops, _legacy_C_ops from paddle.fluid.data_feeder import check_variable_and_dtype from paddle.fluid.dygraph.layer_object_helper import LayerObjectHelper from paddle.fluid.dygraph.nn import InstanceNorm, Conv2D, Conv2DTranspose, BatchNorm from ...common import get_logger from .utils.utils import compute_start_end, get_same_padding, convert_to_list from .layers_base import * __all__ = [ 'SuperConv2D', 'SuperConv2DTranspose', 'SuperSeparableConv2D', 'SuperBatchNorm', 'SuperLinear', 'SuperInstanceNorm', 'SuperGroupConv2D', 'SuperDepthwiseConv2D', 'SuperGroupConv2DTranspose', 'SuperDepthwiseConv2DTranspose', 'SuperLayerNorm', 'SuperEmbedding' ] _logger = get_logger(__name__, level=logging.INFO) ### TODO: if task is elastic width, need to add re_organize_middle_weight in 1x1 conv in MBBlock class SuperConv2D(fluid.dygraph.Conv2D): """ This interface is used to construct a callable object of the ``SuperConv2D`` class. The difference between ```SuperConv2D``` and ```Conv2D``` is: ```SuperConv2D``` need to feed a config dictionary with the format of {'channel', num_of_channel} represents the channels of the outputs, used to change the first dimension of weight and bias, only train the first channels of the weight and bias. Note: the channel in config need to less than first defined. The super convolution2D layer calculates the output based on the input, filter and strides, paddings, dilations, groups parameters. Input and Output are in NCHW format, where N is batch size, C is the number of the feature map, H is the height of the feature map, and W is the width of the feature map. Filter's shape is [MCHW] , where M is the number of output feature map, C is the number of input feature map, H is the height of the filter, and W is the width of the filter. If the groups is greater than 1, C will equal the number of input feature map divided by the groups. Please refer to UFLDL's `convolution `_ for more details. If bias attribution and activation type are provided, bias is added to the output of the convolution, and the corresponding activation function is applied to the final result. For each input :math:`X`, the equation is: .. math:: Out = \\sigma (W \\ast X + b) Where: * :math:`X`: Input value, a ``Tensor`` with NCHW format. * :math:`W`: Filter value, a ``Tensor`` with shape [MCHW] . * :math:`\\ast`: Convolution operation. * :math:`b`: Bias value, a 2-D ``Tensor`` with shape [M, 1]. * :math:`\\sigma`: Activation function. * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different. Example: - Input: Input shape: :math:`(N, C_{in}, H_{in}, W_{in})` Filter shape: :math:`(C_{out}, C_{in}, H_f, W_f)` - Output: Output shape: :math:`(N, C_{out}, H_{out}, W_{out})` Where .. math:: H_{out}&= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]} + 1 \\\\ W_{out}&= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]} + 1 Parameters: num_channels(int): The number of channels in the input image. num_filters(int): The number of filter. It is as same as the output feature map. filter_size (int or tuple): The filter size. If filter_size is a tuple, it must contain two integers, (filter_size_H, filter_size_W). Otherwise, the filter will be a square. candidate_config(dict, optional): Dictionary descripts candidate config of this layer, such as {'kernel_size': (3, 5, 7), 'channel': (4, 6, 8)}, means the kernel size of this layer can be choose from (3, 5, 7), the key of candidate_config only can be 'kernel_size', 'channel' and 'expand_ratio', 'channel' and 'expand_ratio' CANNOT be set at the same time. Default: None. transform_kernel(bool, optional): Whether to use transform matrix to transform a large filter to a small filter. Default: False. stride (int or tuple, optional): The stride size. If stride is a tuple, it must contain two integers, (stride_H, stride_W). Otherwise, the stride_H = stride_W = stride. Default: 1. padding (int or tuple, optional): The padding size. If padding is a tuple, it must contain two integers, (padding_H, padding_W). Otherwise, the padding_H = padding_W = padding. Default: 0. dilation (int or tuple, optional): The dilation size. If dilation is a tuple, it must contain two integers, (dilation_H, dilation_W). Otherwise, the dilation_H = dilation_W = dilation. Default: 1. groups (int, optional): The groups number of the Conv2d Layer. According to grouped convolution in Alex Krizhevsky's Deep CNN paper: when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels. Default: 1. param_attr (ParamAttr, optional): The parameter attribute for learnable weights(Parameter) of conv2d. If it is set to None or one attribute of ParamAttr, conv2d will create ParamAttr as param_attr. If the Initializer of the param_attr is not set, the parameter is initialized with :math:`Normal(0.0, std)`, and the :math:`std` is :math:`(\\frac{2.0 }{filter\\_elem\\_num})^{0.5}`. Default: None. bias_attr (ParamAttr or bool, optional): The attribute for the bias of conv2d. If it is set to False, no bias will be added to the output units. If it is set to None or one attribute of ParamAttr, conv2d will create ParamAttr as bias_attr. If the Initializer of the bias_attr is not set, the bias is initialized zero. Default: None. use_cudnn (bool, optional): Use cudnn kernel or not, it is valid only when the cudnn library is installed. Default: True. act (str, optional): Activation type, if it is set to None, activation is not appended. Default: None. dtype (str, optional): Data type, it can be "float32" or "float64". Default: "float32". Attribute: **weight** (Parameter): the learnable weights of filter of this layer. **bias** (Parameter or None): the learnable bias of this layer. Returns: None Raises: ValueError: if ``use_cudnn`` is not a bool value. Examples: .. code-block:: python from paddle.fluid.dygraph.base import to_variable import paddle.fluid as fluid from paddleslim.core.layers import SuperConv2D import numpy as np data = np.random.uniform(-1, 1, [10, 3, 32, 32]).astype('float32') with fluid.dygraph.guard(): super_conv2d = SuperConv2D(3, 10, 3) config = {'channel': 5} data = to_variable(data) conv = super_conv2d(data, config) """ ### NOTE: filter_size, num_channels and num_filters must be the max of candidate to define a largest network. def __init__(self, num_channels, num_filters, filter_size, candidate_config={}, transform_kernel=False, stride=1, dilation=1, padding=0, groups=None, param_attr=None, bias_attr=None, use_cudnn=True, act=None, dtype='float32'): ### NOTE: padding always is 0, add padding in forward because of kernel size is uncertain super(SuperConv2D, self).__init__( num_channels, num_filters, filter_size, stride, padding, dilation, groups, param_attr, bias_attr, use_cudnn, act, dtype) if isinstance(self._filter_size, int): self._filter_size = convert_to_list(self._filter_size, 2) self.candidate_config = candidate_config if len(candidate_config.items()) != 0: for k, v in candidate_config.items(): candidate_config[k] = list(set(v)) self.ks_set = candidate_config[ 'kernel_size'] if 'kernel_size' in candidate_config else None self.expand_ratio = candidate_config[ 'expand_ratio'] if 'expand_ratio' in candidate_config else None self.channel = candidate_config[ 'channel'] if 'channel' in candidate_config else None self.base_channel = self._num_filters if self.expand_ratio != None: self.base_channel = int(self._num_filters / max(self.expand_ratio)) self.transform_kernel = transform_kernel if self.ks_set != None: self.ks_set.sort() if self.transform_kernel != False: scale_param = dict() ### create parameter to transform kernel for i in range(len(self.ks_set) - 1): ks_small = self.ks_set[i] ks_large = self.ks_set[i + 1] param_name = '%dto%d_matrix' % (ks_large, ks_small) ks_t = ks_small**2 scale_param[param_name] = self.create_parameter( attr=fluid.ParamAttr( name=self._full_name + param_name, initializer=fluid.initializer.NumpyArrayInitializer( np.eye(ks_t))), shape=(ks_t, ks_t), dtype=self._dtype) for name, param in scale_param.items(): setattr(self, name, param) def get_active_filter(self, in_nc, out_nc, kernel_size): ### Unsupport for asymmetric kernels if self._filter_size[0] != self._filter_size[1]: return self.weight[:out_nc, :in_nc, :, :] start, end = compute_start_end(self._filter_size[0], kernel_size) ### if NOT transform kernel, intercept a center filter with kernel_size from largest filter filters = self.weight[:out_nc, :in_nc, start:end, start:end] if self.transform_kernel != False and kernel_size < self._filter_size[ 0]: ### if transform kernel, then use matrix to transform start_filter = self.weight[:out_nc, :in_nc, :, :] for i in range(len(self.ks_set) - 1, 0, -1): src_ks = self.ks_set[i] if src_ks <= kernel_size: break target_ks = self.ks_set[i - 1] start, end = compute_start_end(src_ks, target_ks) _input_filter = start_filter[:, :, start:end, start:end] _input_filter = fluid.layers.reshape( _input_filter, shape=[(_input_filter.shape[0] * _input_filter.shape[1]), -1]) _tmp_filter = _varbase_creator(dtype=_input_filter.dtype) if _non_static_mode(): _legacy_C_ops.matmul(_input_filter, self.__getattr__('%dto%d_matrix' % (src_ks, target_ks)), _tmp_filter, 'transpose_X', False, 'transpose_Y', False, "alpha", 1) _tmp_filter = fluid.layers.reshape( _tmp_filter, shape=[ filters.shape[0], filters.shape[1], target_ks, target_ks ]) start_filter = _tmp_filter filters = start_filter return filters def get_groups_in_out_nc(self, in_nc, out_nc): if self._groups == 1 or self._groups == None: ### standard conv return self._groups, in_nc, out_nc elif self._groups == self._num_channels: ### depthwise convolution if in_nc != out_nc: _logger.debug( "input channel and output channel in depthwise conv is different, change output channel to input channel! origin channel:(in_nc {}, out_nc {}): ". format(in_nc, out_nc)) groups = in_nc out_nc = in_nc return groups, in_nc, out_nc else: ### groups convolution ### conv: weight: (Cout, Cin/G, Kh, Kw) groups = self._groups in_nc = int(in_nc // groups) return groups, in_nc, out_nc def forward(self, input, kernel_size=None, expand_ratio=None, channel=None): self.cur_config = { 'kernel_size': kernel_size, 'expand_ratio': expand_ratio, 'channel': channel } in_nc = int(input.shape[1]) assert ( expand_ratio == None or channel == None ), "expand_ratio and channel CANNOT be NOT None at the same time." if expand_ratio != None: out_nc = int(expand_ratio * self.base_channel) elif channel != None: out_nc = int(channel) else: out_nc = self._num_filters ks = int(self._filter_size[0]) if kernel_size == None else int( kernel_size) if kernel_size is not None and self._filter_size[ 0] != self._filter_size[1]: _logger.error("Searching for asymmetric kernels is NOT supported") groups, weight_in_nc, weight_out_nc = self.get_groups_in_out_nc(in_nc, out_nc) weight = self.get_active_filter(weight_in_nc, weight_out_nc, ks) if kernel_size != None or 'kernel_size' in self.candidate_config.keys(): padding = convert_to_list(get_same_padding(ks), 2) else: padding = self._padding if self._l_type == 'conv2d': attrs = ('strides', self._stride, 'paddings', padding, 'dilations', self._dilation, 'groups', groups if groups else 1, 'use_cudnn', self._use_cudnn) if in_dygraph_mode(): out = _C_ops.conv2d( input, weight, self._stride, padding, "EXPLICIT", groups if groups else 1, self._dilation, "NCHW", False, -1, False) elif _in_legacy_dygraph(): out = _legacy_C_ops.conv2d(input, weight, *attrs) elif self._l_type == 'depthwise_conv2d': attrs = ('strides', self._stride, 'paddings', padding, 'dilations', self._dilation, 'groups', groups if groups else self._groups, 'use_cudnn', self._use_cudnn) out = core.ops.depthwise_conv2d(input, weight, *attrs) else: raise ValueError("conv type error") pre_bias = out out_nc = int(pre_bias.shape[1]) if self.bias is not None: bias = self.bias[:out_nc] pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1) else: pre_act = pre_bias return dygraph_utils._append_activation_in_dygraph(pre_act, self._act) class SuperGroupConv2D(SuperConv2D): def get_groups_in_out_nc(self, in_nc, out_nc): ### groups convolution ### conv: weight: (Cout, Cin/G, Kh, Kw) groups = self._groups in_nc = int(in_nc // groups) return groups, in_nc, out_nc class SuperDepthwiseConv2D(SuperConv2D): ### depthwise convolution def get_groups_in_out_nc(self, in_nc, out_nc): if in_nc != out_nc: _logger.debug( "input channel and output channel in depthwise conv is different, change output channel to input channel! origin channel:(in_nc {}, out_nc {}): ". format(in_nc, out_nc)) groups = in_nc out_nc = in_nc return groups, in_nc, out_nc class SuperConv2DTranspose(fluid.dygraph.Conv2DTranspose): """ This interface is used to construct a callable object of the ``SuperConv2DTranspose`` class. The difference between ```SuperConv2DTranspose``` and ```Conv2DTranspose``` is: ```SuperConv2DTranspose``` need to feed a config dictionary with the format of {'channel', num_of_channel} represents the channels of the outputs, used to change the first dimension of weight and bias, only train the first channels of the weight and bias. Note: the channel in config need to less than first defined. The super convolution2D transpose layer calculates the output based on the input, filter, and dilations, strides, paddings. Input and output are in NCHW format. Where N is batch size, C is the number of feature map, H is the height of the feature map, and W is the width of the feature map. Filter's shape is [MCHW] , where M is the number of input feature map, C is the number of output feature map, H is the height of the filter, and W is the width of the filter. If the groups is greater than 1, C will equal the number of input feature map divided by the groups. If bias attribution and activation type are provided, bias is added to the output of the convolution, and the corresponding activation function is applied to the final result. The details of convolution transpose layer, please refer to the following explanation and references `conv2dtranspose `_ . For each input :math:`X`, the equation is: .. math:: Out = \\sigma (W \\ast X + b) Where: * :math:`X`: Input value, a ``Tensor`` with NCHW format. * :math:`W`: Filter value, a ``Tensor`` with shape [MCHW] . * :math:`\\ast`: Convolution operation. * :math:`b`: Bias value, a 2-D ``Tensor`` with shape [M, 1]. * :math:`\\sigma`: Activation function. * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different. Example: - Input: Input shape: :math:`(N, C_{in}, H_{in}, W_{in})` Filter shape: :math:`(C_{in}, C_{out}, H_f, W_f)` - Output: Output shape: :math:`(N, C_{out}, H_{out}, W_{out})` Where .. math:: H^\\prime_{out} &= (H_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (H_f - 1) + 1 \\\\ W^\\prime_{out} &= (W_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (W_f - 1) + 1 \\\\ H_{out} &\\in [ H^\\prime_{out}, H^\\prime_{out} + strides[0] ) \\\\ W_{out} &\\in [ W^\\prime_{out}, W^\\prime_{out} + strides[1] ) Parameters: num_channels(int): The number of channels in the input image. num_filters(int): The number of the filter. It is as same as the output feature map. filter_size(int or tuple): The filter size. If filter_size is a tuple, it must contain two integers, (filter_size_H, filter_size_W). Otherwise, the filter will be a square. candidate_config(dict, optional): Dictionary descripts candidate config of this layer, such as {'kernel_size': (3, 5, 7), 'channel': (4, 6, 8)}, means the kernel size of this layer can be choose from (3, 5, 7), the key of candidate_config only can be 'kernel_size', 'channel' and 'expand_ratio', 'channel' and 'expand_ratio' CANNOT be set at the same time. Default: None. transform_kernel(bool, optional): Whether to use transform matrix to transform a large filter to a small filter. Default: False. output_size(int or tuple, optional): The output image size. If output size is a tuple, it must contain two integers, (image_H, image_W). None if use filter_size, padding, and stride to calculate output_size. if output_size and filter_size are specified at the same time, They should follow the formula above. Default: None. padding(int or tuple, optional): The padding size. If padding is a tuple, it must contain two integers, (padding_H, padding_W). Otherwise, the padding_H = padding_W = padding. Default: 0. stride(int or tuple, optional): The stride size. If stride is a tuple, it must contain two integers, (stride_H, stride_W). Otherwise, the stride_H = stride_W = stride. Default: 1. dilation(int or tuple, optional): The dilation size. If dilation is a tuple, it must contain two integers, (dilation_H, dilation_W). Otherwise, the dilation_H = dilation_W = dilation. Default: 1. groups(int, optional): The groups number of the Conv2d transpose layer. Inspired by grouped convolution in Alex Krizhevsky's Deep CNN paper, in which when group=2, the first half of the filters is only connected to the first half of the input channels, while the second half of the filters is only connected to the second half of the input channels. Default: 1. param_attr (ParamAttr, optional): The parameter attribute for learnable weights(Parameter) of conv2d_transpose. If it is set to None or one attribute of ParamAttr, conv2d_transpose will create ParamAttr as param_attr. If the Initializer of the param_attr is not set, the parameter is initialized with Xavier. Default: None. bias_attr (ParamAttr or bool, optional): The attribute for the bias of conv2d_transpose. If it is set to False, no bias will be added to the output units. If it is set to None or one attribute of ParamAttr, conv2d_transpose will create ParamAttr as bias_attr. If the Initializer of the bias_attr is not set, the bias is initialized zero. Default: None. use_cudnn(bool, optional): Use cudnn kernel or not, it is valid only when the cudnn library is installed. Default: True. act (str, optional): Activation type, if it is set to None, activation is not appended. Default: None. dtype (str, optional): Data type, it can be "float32" or "float64". Default: "float32". Attribute: **weight** (Parameter): the learnable weights of filters of this layer. **bias** (Parameter or None): the learnable bias of this layer. Returns: None Examples: .. code-block:: python import paddle.fluid as fluid from paddleslim.core.layers import SuperConv2DTranspose import numpy as np with fluid.dygraph.guard(): data = np.random.random((3, 32, 32, 5)).astype('float32') config = {'channel': 5 super_convtranspose = SuperConv2DTranspose(num_channels=32, num_filters=10, filter_size=3) ret = super_convtranspose(fluid.dygraph.base.to_variable(data), config) """ def __init__(self, num_channels, num_filters, filter_size, output_size=None, candidate_config={}, transform_kernel=False, stride=1, dilation=1, padding=0, groups=None, param_attr=None, bias_attr=None, use_cudnn=True, act=None, dtype='float32'): super(SuperConv2DTranspose, self).__init__( num_channels, num_filters, filter_size, output_size, padding, stride, dilation, groups, param_attr, bias_attr, use_cudnn, act, dtype) self.candidate_config = candidate_config if len(self.candidate_config.items()) != 0: for k, v in candidate_config.items(): candidate_config[k] = list(set(v)) self.ks_set = candidate_config[ 'kernel_size'] if 'kernel_size' in candidate_config else None if isinstance(self._filter_size, int): self._filter_size = convert_to_list(self._filter_size, 2) self.expand_ratio = candidate_config[ 'expand_ratio'] if 'expand_ratio' in candidate_config else None self.channel = candidate_config[ 'channel'] if 'channel' in candidate_config else None self.base_channel = self._num_filters if self.expand_ratio: self.base_channel = int(self._num_filters / max(self.expand_ratio)) self.transform_kernel = transform_kernel if self.ks_set != None: self.ks_set.sort() if self.transform_kernel != False: scale_param = dict() ### create parameter to transform kernel for i in range(len(self.ks_set) - 1): ks_small = self.ks_set[i] ks_large = self.ks_set[i + 1] param_name = '%dto%d_matrix' % (ks_large, ks_small) ks_t = ks_small**2 scale_param[param_name] = self.create_parameter( attr=fluid.ParamAttr( name=self._full_name + param_name, initializer=fluid.initializer.NumpyArrayInitializer( np.eye(ks_t))), shape=(ks_t, ks_t), dtype=self._dtype) for name, param in scale_param.items(): setattr(self, name, param) def get_active_filter(self, in_nc, out_nc, kernel_size): ### Unsupport for asymmetric kernels if self._filter_size[0] != self._filter_size[1]: return self.weight[:out_nc, :in_nc, :, :] start, end = compute_start_end(self._filter_size[0], kernel_size) filters = self.weight[:in_nc, :out_nc, start:end, start:end] if self.transform_kernel != False and kernel_size < self._filter_size[ 0]: start_filter = self.weight[:in_nc, :out_nc, :, :] for i in range(len(self.ks_set) - 1, 0, -1): src_ks = self.ks_set[i] if src_ks <= kernel_size: break target_ks = self.ks_set[i - 1] start, end = compute_start_end(src_ks, target_ks) _input_filter = start_filter[:, :, start:end, start:end] _input_filter = fluid.layers.reshape( _input_filter, shape=[(_input_filter.shape[0] * _input_filter.shape[1]), -1]) _tmp_filter = _varbase_creator(dtype=_input_filter.dtype) if _non_static_mode(): _legacy_C_ops.matmul(_input_filter, self.__getattr__('%dto%d_matrix' % (src_ks, target_ks)), _tmp_filter, 'transpose_X', False, 'transpose_Y', False, "alpha", 1) _tmp_filter = fluid.layers.reshape( _tmp_filter, shape=[ filters.shape[0], filters.shape[1], target_ks, target_ks ]) start_filter = _tmp_filter filters = start_filter return filters def get_groups_in_out_nc(self, in_nc, out_nc): if self._groups == 1 or self._groups == None: ### standard conv return self._groups, in_nc, out_nc elif self._groups == self._num_channels: ### depthwise convolution if in_nc != out_nc: _logger.debug( "input channel and output channel in depthwise conv is different, change output channel to input channel! origin channel:(in_nc {}, out_nc {}): ". format(in_nc, out_nc)) groups = in_nc out_nc = in_nc return groups, in_nc, out_nc else: ### groups convolution ### groups conv transpose: weight: (Cin, Cout/G, Kh, Kw) groups = self._groups out_nc = int(out_nc // groups) return groups, in_nc, out_nc def forward(self, input, kernel_size=None, expand_ratio=None, channel=None): self.cur_config = { 'kernel_size': kernel_size, 'expand_ratio': expand_ratio, 'channel': channel } in_nc = int(input.shape[1]) assert ( expand_ratio == None or channel == None ), "expand_ratio and channel CANNOT be NOT None at the same time." if expand_ratio != None: out_nc = int(expand_ratio * self.base_channel) elif channel != None: out_nc = int(channel) else: out_nc = self._num_filters ks = int(self._filter_size[0]) if kernel_size == None else int( kernel_size) if kernel_size is not None and self._filter_size[ 0] != self._filter_size[1]: _logger.error("Searching for asymmetric kernels is NOT supported") groups, weight_in_nc, weight_out_nc = self.get_groups_in_out_nc(in_nc, out_nc) weight = self.get_active_filter(weight_in_nc, weight_out_nc, ks) if kernel_size != None or 'kernel_size' in self.candidate_config.keys(): padding = convert_to_list(get_same_padding(ks), 2) else: padding = self._padding if _non_static_mode(): op = getattr(_legacy_C_ops, self._op_type) out = op(input, weight, 'output_size', self._output_size, 'strides', self._stride, 'paddings', padding, 'dilations', self._dilation, 'groups', groups, 'use_cudnn', self._use_cudnn) pre_bias = out out_nc = int(pre_bias.shape[1]) if self.bias is not None: bias = self.bias[:out_nc] pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1) else: pre_act = pre_bias return dygraph_utils._append_activation_in_dygraph( pre_act, act=self._act) class SuperGroupConv2DTranspose(SuperConv2DTranspose): def get_groups_in_out_nc(self, in_nc, out_nc): ### groups convolution ### groups conv transpose: weight: (Cin, Cout/G, Kh, Kw) groups = self._groups out_nc = int(out_nc // groups) return groups, in_nc, out_nc class SuperDepthwiseConv2DTranspose(SuperConv2DTranspose): def get_groups_in_out_nc(self, in_nc, out_nc): if in_nc != out_nc: _logger.debug( "input channel and output channel in depthwise conv transpose is different, change output channel to input channel! origin channel:(in_nc {}, out_nc {}): ". format(in_nc, out_nc)) groups = in_nc out_nc = in_nc return groups, in_nc, out_nc ### NOTE: only search channel, write for GAN-compression, maybe change to SuperDepthwiseConv and SuperConv after. class SuperSeparableConv2D(fluid.dygraph.Layer): """ This interface is used to construct a callable object of the ``SuperSeparableConv2D`` class. The difference between ```SuperSeparableConv2D``` and ```SeparableConv2D``` is: ```SuperSeparableConv2D``` need to feed a config dictionary with the format of {'channel', num_of_channel} represents the channels of the first conv's outputs and the second conv's inputs, used to change the first dimension of weight and bias, only train the first channels of the weight and bias. The architecture of super separable convolution2D op is [Conv2D, norm layer(may be BatchNorm or InstanceNorm), Conv2D]. The first conv is depthwise conv, the filter number is input channel multiply scale_factor, the group is equal to the number of input channel. The second conv is standard conv, which filter size and stride size are 1. Parameters: num_channels(int): The number of channels in the input image. num_filters(int): The number of the second conv's filter. It is as same as the output feature map. filter_size(int or tuple): The first conv's filter size. If filter_size is a tuple, it must contain two integers, (filter_size_H, filter_size_W). Otherwise, the filter will be a square. padding(int or tuple, optional): The first conv's padding size. If padding is a tuple, it must contain two integers, (padding_H, padding_W). Otherwise, the padding_H = padding_W = padding. Default: 0. stride(int or tuple, optional): The first conv's stride size. If stride is a tuple, it must contain two integers, (stride_H, stride_W). Otherwise, the stride_H = stride_W = stride. Default: 1. dilation(int or tuple, optional): The first conv's dilation size. If dilation is a tuple, it must contain two integers, (dilation_H, dilation_W). Otherwise, the dilation_H = dilation_W = dilation. Default: 1. norm_layer(class): The normalization layer between two convolution. Default: InstanceNorm. bias_attr (ParamAttr or bool, optional): The attribute for the bias of convolution. If it is set to False, no bias will be added to the output units. If it is set to None or one attribute of ParamAttr, convolution will create ParamAttr as bias_attr. If the Initializer of the bias_attr is not set, the bias is initialized zero. Default: None. scale_factor(float): The scale factor of the first conv's output channel. Default: 1. use_cudnn(bool, optional): Use cudnn kernel or not, it is valid only when the cudnn library is installed. Default: True. Returns: None """ def __init__(self, num_channels, num_filters, filter_size, candidate_config={}, stride=1, padding=0, dilation=1, norm_layer=InstanceNorm, bias_attr=None, scale_factor=1, use_cudnn=False): super(SuperSeparableConv2D, self).__init__() self.conv = fluid.dygraph.LayerList([ fluid.dygraph.nn.Conv2D( num_channels=num_channels, num_filters=num_channels * scale_factor, filter_size=filter_size, stride=stride, padding=padding, use_cudnn=False, groups=num_channels, bias_attr=bias_attr) ]) self.conv.extend([norm_layer(num_channels * scale_factor)]) self.conv.extend([ fluid.dygraph.nn.Conv2D( num_channels=num_channels * scale_factor, num_filters=num_filters, filter_size=1, stride=1, use_cudnn=use_cudnn, bias_attr=bias_attr) ]) self.candidate_config = candidate_config self.expand_ratio = candidate_config[ 'expand_ratio'] if 'expand_ratio' in candidate_config else None self.base_output_dim = self.conv[0]._num_filters if self.expand_ratio != None: self.base_output_dim = int(self.conv[0]._num_filters / max(self.expand_ratio)) def forward(self, input, expand_ratio=None, channel=None): self.cur_config = {'expand_ratio': expand_ratio, 'channel': channel} in_nc = int(input.shape[1]) assert ( expand_ratio == None or channel == None ), "expand_ratio and channel CANNOT be NOT None at the same time." if expand_ratio != None: out_nc = int(expand_ratio * self.base_output_dim) elif channel != None: out_nc = int(channel) else: out_nc = self.conv[0]._num_filters weight = self.conv[0].weight[:in_nc] ### conv1 if self.conv[0]._l_type == 'conv2d': if in_dygraph_mode(): out = _C_ops.conv2d(input, weight, self.conv[0]._stride, self.conv[0]._padding, "EXPLICIT", in_nc, self.conv[0]._dilation, "NCHW", False, -1, False) elif _in_legacy_dygraph(): attrs = ('strides', self.conv[0]._stride, 'paddings', self.conv[0]._padding, 'dilations', self.conv[0]._dilation, 'groups', in_nc, 'use_cudnn', self.conv[0]._use_cudnn) out = _legacy_C_ops.conv2d(input, weight, *attrs) elif self.conv[0]._l_type == 'depthwise_conv2d': if in_dygraph_mode(): out = _C_ops.depthwise_conv2d( input, weight, self.conv[0]._stride, self.conv[0]._padding, "EXPLICIT", in_nc, self.conv[0]._dilation, "NCHW", False, -1, False, False, self.conv[0]._use_cudnn) elif _in_legacy_dygraph(): attrs = ('strides', self.conv[0]._stride, 'paddings', self.conv[0]._padding, 'dilations', self.conv[0]._dilation, 'groups', in_nc, 'use_cudnn', self.conv[0]._use_cudnn) out = _legacy_C_ops.depthwise_conv2d(input, weight, *attrs) else: raise ValueError("conv type error") pre_bias = out if self.conv[0].bias is not None: bias = self.conv[0].bias[:in_nc] pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1) else: pre_act = pre_bias conv0_out = dygraph_utils._append_activation_in_dygraph( pre_act, self.conv[0]._act) norm_out = self.conv[1](conv0_out) weight = self.conv[2].weight[:out_nc, :in_nc, :, :] if self.conv[2]._l_type == 'conv2d': if in_dygraph_mode(): out = _C_ops.conv2d( input, weight, self.conv[2]._stride, self.conv[2]._padding, "EXPLICIT", self.conv[2]._groups if self.conv[2]._groups else 1, self.conv[2]._dilation, "NCHW", False, -1, False) elif _in_legacy_dygraph(): attrs = ('strides', self.conv[2]._stride, 'paddings', self.conv[2]._padding, 'dilations', self.conv[2]._dilation, 'groups', self.conv[2]._groups if self.conv[2]._groups else 1, 'use_cudnn', self.conv[2]._use_cudnn) out = _legacy_C_ops.conv2d(norm_out, weight, *attrs) elif self.conv[2]._l_type == 'depthwise_conv2d': attrs = ('strides', self.conv[2]._stride, 'paddings', self.conv[2]._padding, 'dilations', self.conv[2]._dilation, 'groups', self.conv[2]._groups, 'use_cudnn', self.conv[2]._use_cudnn) out = core.ops.depthwise_conv2d(norm_out, weight, *attrs) else: raise ValueError("conv type error") pre_bias = out if self.conv[2].bias is not None: bias = self.conv[2].bias[:out_nc] pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1) else: pre_act = pre_bias conv1_out = dygraph_utils._append_activation_in_dygraph( pre_act, self.conv[2]._act) return conv1_out class SuperLinear(fluid.dygraph.Linear): """ """ def __init__(self, input_dim, output_dim, candidate_config={}, param_attr=None, bias_attr=None, act=None, dtype="float32"): super(SuperLinear, self).__init__(input_dim, output_dim, param_attr, bias_attr, act, dtype) self._param_attr = param_attr self._bias_attr = bias_attr self.output_dim = output_dim self.candidate_config = candidate_config self.expand_ratio = candidate_config[ 'expand_ratio'] if 'expand_ratio' in candidate_config else None self.base_output_dim = self.output_dim if self.expand_ratio != None: self.base_output_dim = int(self.output_dim / max(self.expand_ratio)) def forward(self, input, expand_ratio=None, channel=None): self.cur_config = {'expand_ratio': expand_ratio, 'channel': channel} ### weight: (Cin, Cout) in_nc = int(input.shape[-1]) assert ( expand_ratio == None or channel == None ), "expand_ratio and channel CANNOT be NOT None at the same time." if expand_ratio != None: out_nc = int(expand_ratio * self.base_output_dim) elif channel != None: out_nc = int(channel) else: out_nc = self.output_dim weight = self.weight[:in_nc, :out_nc] if self._bias_attr != False: bias = self.bias[:out_nc] use_bias = True pre_bias = _varbase_creator(dtype=input.dtype) if _non_static_mode(): _legacy_C_ops.matmul(input, weight, pre_bias, 'transpose_X', False, 'transpose_Y', False, "alpha", 1) if self._bias_attr != False: pre_act = dygraph_utils._append_bias_in_dygraph( pre_bias, bias, axis=len(input.shape) - 1) else: pre_act = pre_bias return dygraph_utils._append_activation_in_dygraph(pre_act, self._act) class SuperBatchNorm(fluid.dygraph.BatchNorm): """ add comment """ def __init__(self, num_channels, act=None, is_test=False, momentum=0.9, epsilon=1e-05, param_attr=None, bias_attr=None, dtype='float32', data_layout='NCHW', in_place=False, moving_mean_name=None, moving_variance_name=None, do_model_average_for_mean_and_var=True, use_global_stats=False, trainable_statistics=False): super(SuperBatchNorm, self).__init__( num_channels, act, is_test, momentum, epsilon, param_attr, bias_attr, dtype, data_layout, in_place, moving_mean_name, moving_variance_name, do_model_average_for_mean_and_var, use_global_stats, trainable_statistics) def forward(self, input): feature_dim = int(input.shape[1]) weight = self.weight[:feature_dim] bias = self.bias[:feature_dim] mean = self._mean[:feature_dim] variance = self._variance[:feature_dim] mean_out = self._mean variance_out = self._variance mean_out_tmp = mean variance_out_tmp = variance attrs = ("momentum", self._momentum, "epsilon", self._epsilon, "is_test", not self.training, "data_layout", self._data_layout, "use_mkldnn", False, "fuse_with_relu", self._fuse_with_relu, "use_global_stats", self._use_global_stats, 'trainable_statistics', self._trainable_statistics) if in_dygraph_mode(): if feature_dim != self._mean.shape[0]: batch_norm_out, t1, t2, t3, t4, _ = _C_ops.batch_norm( input, weight, bias, mean, variance, self._momentum, self._epsilon, self._data_layout, not self.training, self._use_global_stats, self._trainable_statistics, False) self._mean[:feature_dim] = mean self._variance[:feature_dim] = variance mean_out[:feature_dim] = mean_out_tmp variance_out[:feature_dim] = variance_out_tmp else: batch_norm_out, t1, t2, t3, t4, _ = _C_ops.batch_norm( input, weight, bias, mean, variance, self._momentum, self._epsilon, self._data_layout, not self.training, self._use_global_stats, self._trainable_statistics, False) return batch_norm_out elif _in_legacy_dygraph(): if feature_dim != self._mean.shape[0]: batch_norm_out, t1, t2, t3, t4, _ = _legacy_C_ops.batch_norm( input, weight, bias, mean, variance, None, mean_out_tmp, variance_out_tmp, *attrs) self._mean[:feature_dim].set_value(mean) self._variance[:feature_dim].set_value(variance) mean_out[:feature_dim].set_value(mean_out_tmp) variance_out[:feature_dim].set_value(variance_out_tmp) else: batch_norm_out, t1, t2, t3, t4, _ = _legacy_C_ops.batch_norm( input, weight, bias, self._mean, self._variance, None, mean_out, variance_out, *attrs) return batch_norm_out else: check_variable_and_dtype( input, 'input', ['float16', 'float32', 'float64'], 'BatchNorm') # for static need dict attrs = { "momentum": self._momentum, "epsilon": self._epsilon, "is_test": not self.training, "data_layout": self._data_layout, "use_mkldnn": False, "fuse_with_relu": False, "use_global_stats": self._use_global_stats, "trainable_statistics": self._trainable_statistics, } inputs = { "X": [input], "Scale": [weight], "Bias": [bias], "Mean": [mean], "Variance": [variance] } helper = LayerObjectHelper('batch_norm') param_dtype = input.dtype if input.dtype != 'float16' else 'float32' saved_mean = helper.create_variable_for_type_inference( dtype=param_dtype, stop_gradient=True) saved_variance = helper.create_variable_for_type_inference( dtype=param_dtype, stop_gradient=True) batch_norm_out = helper.create_variable_for_type_inference( input.dtype) outputs = { "Y": [batch_norm_out], "MeanOut": [mean], "VarianceOut": [variance], "SavedMean": [saved_mean], "SavedVariance": [saved_variance] } if self.training or self._trainable_statistics: # reserve_space is only used for training. reserve_space = helper.create_variable_for_type_inference( dtype=input.dtype, stop_gradient=True) outputs["ReserveSpace"] = [reserve_space] helper.append_op( type="batch_norm", inputs=inputs, outputs=outputs, attrs=attrs) return batch_norm_out class SuperInstanceNorm(fluid.dygraph.InstanceNorm): """ """ def __init__(self, num_channels, epsilon=1e-05, param_attr=None, bias_attr=None, dtype='float32'): super(SuperInstanceNorm, self).__init__(num_channels, epsilon, param_attr, bias_attr, dtype) def forward(self, input): feature_dim = int(input.shape[1]) if self._param_attr == False and self._bias_attr == False: scale = None bias = None else: scale = self.scale[:feature_dim] bias = self.bias[:feature_dim] if in_dygraph_mode(): out = _C_ops.instance_norm(input, scale, bias, self._epsilon) return out if _in_legacy_dygraph(): out, _, _ = _legacy_C_ops.instance_norm(input, scale, bias, 'epsilon', self._epsilon) return out class SuperLayerNorm(fluid.dygraph.LayerNorm): def __init__(self, normalized_shape, scale=True, shift=True, epsilon=1e-05, param_attr=None, bias_attr=None, act=None, dtype='float32'): super(SuperLayerNorm, self).__init__(normalized_shape, scale, shift, epsilon, param_attr, bias_attr, act, dtype) def forward(self, input): input_shape = list(input.shape) input_ndim = len(input_shape) normalized_ndim = len(self._normalized_shape) self._begin_norm_axis = input_ndim - normalized_ndim ### TODO(ceci3): fix if normalized_shape is not a single number feature_dim = int(input.shape[-1]) weight = self.weight[:feature_dim] bias = self.bias[:feature_dim] if in_dygraph_mode(): pre_act, _, _, = _C_ops.layer_norm(input, weight, bias, self._epsilon, self._begin_norm_axis, False) elif _in_legacy_dygraph(): pre_act, _, _ = _legacy_C_ops.layer_norm( input, weight, bias, 'epsilon', self._epsilon, 'begin_norm_axis', self._begin_norm_axis) return pre_act class SuperEmbedding(fluid.dygraph.Embedding): def __init__(self, size, candidate_config={}, is_sparse=False, is_distributed=False, padding_idx=None, param_attr=None, dtype='float32'): super(SuperEmbedding, self).__init__(size, is_sparse, is_distributed, padding_idx, param_attr, dtype) self.candidate_config = candidate_config self.expand_ratio = candidate_config[ 'expand_ratio'] if 'expand_ratio' in candidate_config else None self.base_output_dim = self._size[-1] if self.expand_ratio != None: self.base_output_dim = int(self._size[-1] / max(self.expand_ratio)) def forward(self, input, expand_ratio=None, channel=None): assert ( expand_ratio == None or channel == None ), "expand_ratio and channel CANNOT be NOT None at the same time." if expand_ratio != None: out_nc = int(expand_ratio * self.base_output_dim) elif channel != None: out_nc = int(channel) else: out_nc = self._size[-1] weight = self.weight[:, :out_nc] if in_dygraph_mode(): return _C_ops.embedding(input, weight, self._padding_idx, self._is_sparse) elif _in_legacy_dygraph(): return _legacy_C_ops.lookup_table_v2( weight, input, 'is_sparse', self._is_sparse, 'is_distributed', self._is_distributed, 'remote_prefetch', self._remote_prefetch, 'padding_idx', self._padding_idx)