From 70e44732c2c1a2186d26a076c3b3be69b6a91bc4 Mon Sep 17 00:00:00 2001 From: wangyang59 Date: Tue, 25 Oct 2016 13:55:40 -0700 Subject: [PATCH] added convTrans test and python components --- .gitignore | 2 + paddle/gserver/tests/CMakeLists.txt | 8 + paddle/gserver/tests/test_ConvTrans.cpp | 139 ++++++++++++++++++ python/paddle/trainer/config_parser.py | 95 ++++++++++++ .../paddle/trainer_config_helpers/layers.py | 123 ++++++++++++++++ 5 files changed, 367 insertions(+) create mode 100644 paddle/gserver/tests/test_ConvTrans.cpp diff --git a/.gitignore b/.gitignore index 65ba217de37..ee8489c1d71 100644 --- a/.gitignore +++ b/.gitignore @@ -5,4 +5,6 @@ build/ .vscode .idea .project +.cproject .pydevproject +Makefile diff --git a/paddle/gserver/tests/CMakeLists.txt b/paddle/gserver/tests/CMakeLists.txt index 26ee2b3aae6..0651d0b4733 100644 --- a/paddle/gserver/tests/CMakeLists.txt +++ b/paddle/gserver/tests/CMakeLists.txt @@ -26,6 +26,14 @@ add_unittest_without_exec(test_ActivationGrad TestUtil.cpp) add_test(NAME test_ActivationGrad COMMAND test_ActivationGrad) +################# test_ConvTrans ####################### +add_unittest_without_exec(test_ConvTrans + test_ConvTrans.cpp + LayerGradUtil.cpp + TestUtil.cpp) + +add_test(NAME test_ConvTrans + COMMAND test_ConvTrans) ################## test_Evaluator ####################### add_unittest(test_Evaluator diff --git a/paddle/gserver/tests/test_ConvTrans.cpp b/paddle/gserver/tests/test_ConvTrans.cpp new file mode 100644 index 00000000000..e7cbe2614fa --- /dev/null +++ b/paddle/gserver/tests/test_ConvTrans.cpp @@ -0,0 +1,139 @@ +/* Copyright (c) 2016 Baidu, Inc. 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. */ + +#include +#include +#include +#include "paddle/gserver/layers/DataLayer.h" +#include "ModelConfig.pb.h" +#include "paddle/trainer/Trainer.h" +#include "paddle/utils/GlobalConstants.h" +#include "paddle/gserver/layers/ExpandConvTransLayer.h" + +#include "TestUtil.h" +#include "LayerGradUtil.h" + +using namespace paddle; // NOLINT +using namespace std; // NOLINT + +P_DECLARE_bool(use_gpu); +P_DECLARE_int32(gpu_id); +P_DECLARE_double(checkgrad_eps); +P_DECLARE_bool(thread_local_rand_use_global_seed); +P_DECLARE_bool(prev_batch_state); + +TEST(Layer, convTransLayerFwd) { + TestConfig configt; + configt.biasSize = 3; + configt.layerConfig.set_type("exconvt"); + configt.layerConfig.set_num_filters(3); + configt.layerConfig.set_partial_sum(1); + configt.layerConfig.set_shared_biases(true); + + configt.inputDefs.push_back({INPUT_DATA, "layer_0", 1024, 288}); + LayerInputConfig* input = configt.layerConfig.add_inputs(); + ConvConfig* conv = input->mutable_conv_conf(); + conv->set_filter_size(2); + conv->set_filter_size_y(3); + conv->set_channels(16); + conv->set_padding(0); + conv->set_padding_y(1); + conv->set_stride(2); + conv->set_stride_y(2); + conv->set_groups(1); + conv->set_filter_channels(3 / conv->groups()); + conv->set_img_size(16); + conv->set_output_x( + (2 * conv->padding() + conv->img_size() - conv->filter_size()) / + ((float)conv->stride()) + + 1.5); + + configt.layerConfig.set_size(conv->img_size() * conv->img_size() * + configt.layerConfig.num_filters()); + configt.layerConfig.set_name("convTrans"); + + // data layer initialize + std::vector dataLayers; + LayerMap layerMap; + vector datas; + initDataLayer(configt, &dataLayers, &datas, &layerMap, "convTrans", + 100, false, useGpu); + // test layer initialize + std::vector parameters; + LayerPtr convtLayer; + initTestLayer(configt, &layerMap, ¶meters, &convtLayer); + convtLayer->getBiasParameter()->zeroMem(); + convtLayer->forward(PASS_GC); + + TestConfig config; + config.biasSize = 16; + config.layerConfig.set_type("exconv"); + config.layerConfig.set_num_filters(16); + config.layerConfig.set_partial_sum(1); + config.layerConfig.set_shared_biases(true); + + config.inputDefs.push_back({INPUT_DATA, "layer_1", 768, 288}); + input = config.layerConfig.add_inputs(); + conv = input->mutable_conv_conf(); + conv->set_filter_size(2); + conv->set_filter_size_y(3); + conv->set_channels(3); + conv->set_padding(0); + conv->set_padding_y(1); + conv->set_stride(2); + conv->set_stride_y(2); + conv->set_groups(1); + conv->set_filter_channels(conv->channels() / conv->groups()); + conv->set_img_size(16); + conv->set_output_x( + (2 * conv->padding() + conv->img_size() - conv->filter_size()) / + ((float)conv->stride()) + + 1.5); + config.layerConfig.set_size(conv->output_x() * conv->output_x() * + config.layerConfig.num_filters()); + config.layerConfig.set_name("conv"); + + // data layer initialize + std::vector dataLayers2; + LayerMap layerMap2; + vector datas2; + initDataLayer(config, &dataLayers2, &datas2, &layerMap2, "conv", + 100, false, useGpu); + // test layer initialize + std::vector parameters2; + LayerPtr convLayer; + initTestLayer(config, &layerMap2, ¶meters2, &convLayer); + + convLayer->getBiasParameter()->zeroMem(); + convLayer->getParameters()[0]->getBuf(PARAMETER_VALUE)->copyFrom( + *(convtLayer->getParameters()[0]->getBuf(PARAMETER_VALUE))); + + convLayer->forward(PASS_GC); + convLayer->getOutput().grad->copyFrom(*(dataLayers[0]->getOutputValue())); + + vector callbackFlags(parameters2.size(), 0); + auto callback = [&](Parameter* para) { ++callbackFlags[para->getID()]; }; + convLayer->backward(callback); + + checkMatrixEqual(convtLayer->getOutputValue(), + dataLayers2[0]->getOutputGrad()); +} + +int main(int argc, char** argv) { + testing::InitGoogleTest(&argc, argv); + initMain(argc, argv); + FLAGS_thread_local_rand_use_global_seed = true; + srand(1); + return RUN_ALL_TESTS(); +} diff --git a/python/paddle/trainer/config_parser.py b/python/paddle/trainer/config_parser.py index 73631602a92..2d28b34999c 100644 --- a/python/paddle/trainer/config_parser.py +++ b/python/paddle/trainer/config_parser.py @@ -1106,6 +1106,37 @@ def parse_conv(conv, input_layer_name, conv_conf): conv_conf.padding, conv_conf.stride, conv_conf.caffe_mode) + +def parse_convt(conv, input_layer_name, conv_conf): + conv_conf.filter_size = conv.filter_size + conv_conf.filter_size_y = conv.filter_size_y + conv_conf.channels = conv.channels + conv_conf.padding = conv.padding + conv_conf.padding_y = conv.padding_y + conv_conf.stride = conv.stride + conv_conf.stride_y = conv.stride_y + conv_conf.groups = conv.groups + conv_conf.filter_channels = conv.channels / conv.groups + conv_conf.caffe_mode = conv.caffe_mode + + outputSize = g_layer_map[input_layer_name].size / conv.channels + print('channels=%d size=%d'%(conv.channels, + g_layer_map[input_layer_name].size)) + conv_conf.output_x = int(outputSize ** 0.5) + config_assert((conv_conf.output_x ** 2) == outputSize, + ("Input layer %s: Incorrect input image size %d for input " + + "image pixels %d") + % (input_layer_name, conv_conf.img_size, img_pixels)) + if conv.caffe_mode: + conv_conf.img_size = \ + (conv_conf.output_x - 1) * conv.stride \ + + conv.filter_size - 2 * conv.padding + else: + conv_conf.img_size = \ + (conv_conf.output_x - 1) * conv.stride \ + + conv.filter_size - 2 * conv.padding + 1 + + def parse_block_expand(block_expand, input_layer_name, block_expand_conf): block_expand_conf.channels = block_expand.channels block_expand_conf.stride_x = block_expand.stride_x @@ -1612,6 +1643,70 @@ class ConvLayer(ConvLayerBase): class ConvLayer(ConvLayerBase): layer_type = 'cudnn_conv' + +@config_layer('convt') +class ConvTransLayerBase(LayerBase): + layer_type = 'convt' + def __init__( + self, + name, + inputs=[], + bias=True, + num_filters=None, + shared_biases=False, + **xargs): + super(ConvLayerBase, self).__init__( + name, self.layer_type, 0, inputs=inputs, **xargs) + + if num_filters is not None: + self.config.num_filters = num_filters + + use_gpu = int(g_command_config_args.get("use_gpu", 0)) + parallel_nn = int(g_command_config_args.get("parallel_nn", 0)) + + # Automatically select cudnn_type for GPU and exconv for CPU + # if set type=conv, but still reserve the way user specify + # exconv or cudnn_conv manually. + if self.layer_type == "cudnn_convt": + config_assert(use_gpu, "cudnn_convt only support GPU") + + if (use_gpu == 1 and self.layer_type != "exconvt" and + (parallel_nn == 0 or self.config.device > -1)): + self.layer_type = "cudnn_convt" + else: + self.layer_type = "exconvt" + # need to specify layer in config + self.config.type = self.layer_type + + if shared_biases is not None: + self.config.shared_biases = shared_biases + + for input_index in xrange(len(self.inputs)): + input_layer = self.get_input_layer(input_index) + parse_convt( + self.inputs[input_index].conv, + input_layer.name, + self.config.inputs[input_index].conv_conf) + conv_conf = self.config.inputs[input_index].conv_conf + psize = self.calc_parameter_size(conv_conf) + print("output size for %s is %d " % (name, conv_conf.output_x)) + self.create_input_parameter(input_index, psize) + self.set_layer_size( + (conv_conf.img_size ** 2) * self.config.num_filters) + + psize = self.config.size + if shared_biases: + psize = self.config.num_filters + self.create_bias_parameter(bias, psize, [psize, 1]) + + def calc_parameter_size(self, conv_conf): + return conv_conf.channels() * conv_conf.filter_channels \ + * (conv_conf.filter_size * conv_conf.filter_size_y) + +@config_layer('exconvt') +class ConvTransLayer(ConvTransLayerBase): + layer_type = 'exconvt' + @config_layer('norm') class NormLayer(LayerBase): def __init__( diff --git a/python/paddle/trainer_config_helpers/layers.py b/python/paddle/trainer_config_helpers/layers.py index 49f0ff3289d..853df8b8370 100644 --- a/python/paddle/trainer_config_helpers/layers.py +++ b/python/paddle/trainer_config_helpers/layers.py @@ -78,6 +78,7 @@ class LayerType(object): COSINE_SIM = 'cos' HSIGMOID = 'hsigmoid' CONV_LAYER = "conv" + CONVTRANS_LAYER = "convt" POOL_LAYER = "pool" BATCH_NORM_LAYER = 'batch_norm' NORM_LAYER = 'norm' @@ -1625,6 +1626,128 @@ def img_conv_layer(input, filter_size, num_filters, return LayerOutput(name, LayerType.CONV_LAYER, parents=[input], activation=act, num_filters=num_filters) +@wrap_name_default("convt") +@wrap_param_attr_default() +@wrap_bias_attr_default() +@wrap_act_default(act=ReluActivation()) +@layer_support(DROPOUT) +def img_convTrans_layer(input, filter_size, num_filters, + name=None, num_channels=None, + act=None, groups=1, stride=1, padding=0, bias_attr=None, + param_attr=None, shared_biases=True, layer_attr=None, + filter_size_y=None, stride_y=None, padding_y=None): + """ + Convolution Transpose (deconv) layer for image. Paddle only support square + input currently and thus input image's width equals height. + + The details of convolution transpose layer, + please refer to the following explanation and references therein + `_ . + + The num_channel means input image's channel number. It may be 1 or 3 when + input is raw pixels of image(mono or RGB), or it may be the previous layer's + num_filters * num_group. + + There are several group of filter in PaddlePaddle implementation. + Each group will process some channel of the inputs. For example, if an input + num_channel = 256, group = 4, num_filter=32, the PaddlePaddle will create + 32*4 = 128 filters to process inputs. The channels will be split into 4 + pieces. First 256/4 = 64 channels will process by first 32 filters. The + rest channels will be processed by rest group of filters. + + :param name: Layer name. + :type name: basestring + :param input: Layer Input. + :type input: LayerOutput + :param filter_size: The x dimension of a filter kernel. Or input a tuple for + two image dimension. + :type filter_size: int|tuple|list + :param filter_size_y: The y dimension of a filter kernel. Since PaddlePaddle + currently supports rectangular filters, the filter's + shape will be (filter_size, filter_size_y). + :type filter_size_y: int|None + :param num_filters: Each filter group's number of filter + :param act: Activation type. Default is tanh + :type act: BaseActivation + :param groups: Group size of filters. + :type groups: int + :param stride: The x dimension of the stride. Or input a tuple for two image + dimension. + :type stride: int|tuple|list + :param stride_y: The y dimension of the stride. + :type stride_y: int + :param padding: The x dimension of the padding. Or input a tuple for two + image dimension + :type padding: int|tuple|list + :param padding_y: The y dimension of the padding. + :type padding_y: int + :param bias_attr: Convolution bias attribute. None means default bias. + False means no bias. + :type bias_attr: ParameterAttribute|False + :param num_channels: number of input channels. If None will be set + automatically from previous output. + :type num_channels: int + :param param_attr: Convolution param attribute. None means default attribute + :type param_attr: ParameterAttribute + :param shared_biases: Is biases will be shared between filters or not. + :type shared_biases: bool + :param layer_attr: Layer Extra Attribute. + :type layer_attr: ExtraLayerAttribute + :return: LayerOutput object. + :rtype: LayerOutput + """ + if num_channels is None: + assert input.num_filters is not None + num_channels = input.num_filters + + if filter_size_y is None: + if isinstance(filter_size, collections.Sequence): + assert len(filter_size) == 2 + filter_size, filter_size_y = filter_size + else: + filter_size_y = filter_size + + if stride_y is None: + if isinstance(stride, collections.Sequence): + assert len(stride) == 2 + stride, stride_y = stride + else: + stride_y = stride + + if padding_y is None: + if isinstance(padding, collections.Sequence): + assert len(padding) == 2 + padding, padding_y = padding + else: + padding_y = padding + + if param_attr.attr.get('initial_smart'): + # special initial for conv layers. + init_w = (2.0 / (filter_size ** 2 * num_channels)) ** 0.5 + param_attr.attr["initial_mean"] = 0.0 + param_attr.attr["initial_std"] = init_w + param_attr.attr["initial_strategy"] = 0 + param_attr.attr["initial_smart"] = False + Layer( + name=name, + inputs=Input(input.name, conv=Conv( + filter_size=filter_size, padding=padding, stride=stride, + channels=num_channels, groups=groups, + filter_size_y=filter_size_y, padding_y=padding_y, + stride_y=stride_y), + **param_attr.attr), + active_type=act.name, + num_filters=num_filters, + bias=ParamAttr.to_bias(bias_attr), + shared_biases=shared_biases, + type=LayerType.CONVTRANS_LAYER, + **ExtraLayerAttribute.to_kwargs(layer_attr) + ) + return LayerOutput(name, LayerType.CONVTRANS_LAYER, parents=[input], + activation=act, num_filters=num_filters) + + @wrap_name_default("pool") @layer_support() -- GitLab