emailweixu
8 years ago
committed by
GitHub
commit
95ef1af2be
@ -0,0 +1,11 @@
|
||||
output/
|
||||
uniform_params/
|
||||
cifar_params/
|
||||
mnist_params/
|
||||
*.png
|
||||
.pydevproject
|
||||
.project
|
||||
*.log
|
||||
*.pyc
|
||||
data/mnist_data/
|
||||
data/cifar-10-batches-py/
|
||||
@ -0,0 +1,13 @@
|
||||
# Generative Adversarial Networks (GAN)
|
||||
|
||||
This demo implements GAN training described in the original GAN paper (https://arxiv.org/abs/1406.2661) and DCGAN (https://arxiv.org/abs/1511.06434).
|
||||
|
||||
The general training procedures are implemented in gan_trainer.py. The neural network configurations are specified in gan_conf.py (for synthetic data) and gan_conf_image.py (for image data).
|
||||
|
||||
In order to run the model, first download the corresponding data by running the shell script in ./data.
|
||||
Then you can run the command below. The flag -d specifies the training data (cifar, mnist or uniform) and flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu).
|
||||
|
||||
$python gan_trainer.py -d cifar --use_gpu 1
|
||||
|
||||
The generated images will be stored in ./cifar_samples/
|
||||
The corresponding models will be stored in ./cifar_params/
|
||||
@ -0,0 +1,18 @@
|
||||
# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
set -e
|
||||
wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
|
||||
tar zxf cifar-10-python.tar.gz
|
||||
rm cifar-10-python.tar.gz
|
||||
|
||||
@ -0,0 +1,19 @@
|
||||
#!/usr/bin/env sh
|
||||
# This script downloads the mnist data and unzips it.
|
||||
set -e
|
||||
DIR="$( cd "$(dirname "$0")" ; pwd -P )"
|
||||
rm -rf "$DIR/mnist_data"
|
||||
mkdir "$DIR/mnist_data"
|
||||
cd "$DIR/mnist_data"
|
||||
|
||||
echo "Downloading..."
|
||||
|
||||
for fname in train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte
|
||||
do
|
||||
if [ ! -e $fname ]; then
|
||||
wget --no-check-certificate http://yann.lecun.com/exdb/mnist/${fname}.gz
|
||||
gunzip ${fname}.gz
|
||||
fi
|
||||
done
|
||||
|
||||
|
||||
@ -0,0 +1,134 @@
|
||||
# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
|
||||
#
|
||||
# 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.
|
||||
from paddle.trainer_config_helpers import *
|
||||
|
||||
mode = get_config_arg("mode", str, "generator")
|
||||
assert mode in set(["generator",
|
||||
"discriminator",
|
||||
"generator_training",
|
||||
"discriminator_training"])
|
||||
|
||||
is_generator_training = mode == "generator_training"
|
||||
is_discriminator_training = mode == "discriminator_training"
|
||||
is_generator = mode == "generator"
|
||||
is_discriminator = mode == "discriminator"
|
||||
|
||||
# The network structure below follows the ref https://arxiv.org/abs/1406.2661
|
||||
# Here we used two hidden layers and batch_norm
|
||||
|
||||
print('mode=%s' % mode)
|
||||
# the dim of the noise (z) as the input of the generator network
|
||||
noise_dim = 10
|
||||
# the dim of the hidden layer
|
||||
hidden_dim = 10
|
||||
# the dim of the generated sample
|
||||
sample_dim = 2
|
||||
|
||||
settings(
|
||||
batch_size=128,
|
||||
learning_rate=1e-4,
|
||||
learning_method=AdamOptimizer(beta1=0.5)
|
||||
)
|
||||
|
||||
def discriminator(sample):
|
||||
"""
|
||||
discriminator ouputs the probablity of a sample is from generator
|
||||
or real data.
|
||||
The output has two dimenstional: dimension 0 is the probablity
|
||||
of the sample is from generator and dimension 1 is the probabblity
|
||||
of the sample is from real data.
|
||||
"""
|
||||
param_attr = ParamAttr(is_static=is_generator_training)
|
||||
bias_attr = ParamAttr(is_static=is_generator_training,
|
||||
initial_mean=1.0,
|
||||
initial_std=0)
|
||||
|
||||
hidden = fc_layer(input=sample, name="dis_hidden", size=hidden_dim,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=ReluActivation())
|
||||
|
||||
hidden2 = fc_layer(input=hidden, name="dis_hidden2", size=hidden_dim,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=LinearActivation())
|
||||
|
||||
hidden_bn = batch_norm_layer(hidden2,
|
||||
act=ReluActivation(),
|
||||
name="dis_hidden_bn",
|
||||
bias_attr=bias_attr,
|
||||
param_attr=ParamAttr(is_static=is_generator_training,
|
||||
initial_mean=1.0,
|
||||
initial_std=0.02),
|
||||
use_global_stats=False)
|
||||
|
||||
return fc_layer(input=hidden_bn, name="dis_prob", size=2,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=SoftmaxActivation())
|
||||
|
||||
def generator(noise):
|
||||
"""
|
||||
generator generates a sample given noise
|
||||
"""
|
||||
param_attr = ParamAttr(is_static=is_discriminator_training)
|
||||
bias_attr = ParamAttr(is_static=is_discriminator_training,
|
||||
initial_mean=1.0,
|
||||
initial_std=0)
|
||||
|
||||
hidden = fc_layer(input=noise,
|
||||
name="gen_layer_hidden",
|
||||
size=hidden_dim,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=ReluActivation())
|
||||
|
||||
hidden2 = fc_layer(input=hidden, name="gen_hidden2", size=hidden_dim,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=LinearActivation())
|
||||
|
||||
hidden_bn = batch_norm_layer(hidden2,
|
||||
act=ReluActivation(),
|
||||
name="gen_layer_hidden_bn",
|
||||
bias_attr=bias_attr,
|
||||
param_attr=ParamAttr(is_static=is_discriminator_training,
|
||||
initial_mean=1.0,
|
||||
initial_std=0.02),
|
||||
use_global_stats=False)
|
||||
|
||||
return fc_layer(input=hidden_bn,
|
||||
name="gen_layer1",
|
||||
size=sample_dim,
|
||||
bias_attr=bias_attr,
|
||||
param_attr=param_attr,
|
||||
act=LinearActivation())
|
||||
|
||||
if is_generator_training:
|
||||
noise = data_layer(name="noise", size=noise_dim)
|
||||
sample = generator(noise)
|
||||
|
||||
if is_discriminator_training:
|
||||
sample = data_layer(name="sample", size=sample_dim)
|
||||
|
||||
if is_generator_training or is_discriminator_training:
|
||||
label = data_layer(name="label", size=1)
|
||||
prob = discriminator(sample)
|
||||
cost = cross_entropy(input=prob, label=label)
|
||||
classification_error_evaluator(input=prob, label=label, name=mode+'_error')
|
||||
outputs(cost)
|
||||
|
||||
if is_generator:
|
||||
noise = data_layer(name="noise", size=noise_dim)
|
||||
outputs(generator(noise))
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,120 @@
|
||||
/* 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 <gtest/gtest.h>
|
||||
#include <vector>
|
||||
#include <string>
|
||||
#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 that the batchNormLayer can be followed by a ConvLayer
|
||||
TEST(Layer, batchNorm) {
|
||||
FLAGS_use_gpu = false;
|
||||
TestConfig configBN;
|
||||
const int CHANNELS = 6272;
|
||||
const int IMG_SIZE = 1;
|
||||
configBN.layerConfig.set_type("batch_norm");
|
||||
configBN.layerConfig.set_name("bn");
|
||||
configBN.layerConfig.set_size(CHANNELS * IMG_SIZE * IMG_SIZE);
|
||||
configBN.layerConfig.set_active_type("relu");
|
||||
configBN.biasSize = CHANNELS;
|
||||
configBN.inputDefs.push_back({INPUT_DATA, "layer_0",
|
||||
/* dim= */ IMG_SIZE * IMG_SIZE * CHANNELS,
|
||||
/* paraSize= */ CHANNELS});
|
||||
|
||||
configBN.inputDefs.push_back({INPUT_DATA, "layer_1_running_mean",
|
||||
1, CHANNELS});
|
||||
configBN.inputDefs.back().isStatic = true;
|
||||
configBN.inputDefs.push_back({INPUT_DATA, "layer_2_running_var",
|
||||
1, CHANNELS});
|
||||
configBN.inputDefs.back().isStatic = true;
|
||||
|
||||
LayerInputConfig* input = configBN.layerConfig.add_inputs();
|
||||
configBN.layerConfig.add_inputs();
|
||||
configBN.layerConfig.add_inputs();
|
||||
|
||||
ImageConfig* img_conf = input->mutable_image_conf();
|
||||
img_conf->set_channels(CHANNELS);
|
||||
img_conf->set_img_size(IMG_SIZE);
|
||||
|
||||
// Setting up conv-layer config
|
||||
TestConfig config;
|
||||
config.biasSize = 64;
|
||||
config.layerConfig.set_type("exconv");
|
||||
config.layerConfig.set_num_filters(64);
|
||||
config.layerConfig.set_partial_sum(1);
|
||||
config.layerConfig.set_shared_biases(true);
|
||||
|
||||
config.inputDefs.push_back({INPUT_DATA, "bn", 6272, 204800});
|
||||
input = config.layerConfig.add_inputs();
|
||||
ConvConfig* conv = input->mutable_conv_conf();
|
||||
conv->set_filter_size(5);
|
||||
conv->set_filter_size_y(5);
|
||||
conv->set_channels(128);
|
||||
conv->set_padding(1);
|
||||
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(7);
|
||||
conv->set_output_x(3);
|
||||
config.layerConfig.set_size(conv->output_x() * conv->output_x() *
|
||||
config.layerConfig.num_filters());
|
||||
config.layerConfig.set_name("conv");
|
||||
|
||||
// data layer initialize
|
||||
std::vector<DataLayerPtr> dataLayers;
|
||||
LayerMap layerMap;
|
||||
vector<Argument> datas;
|
||||
initDataLayer(configBN, &dataLayers, &datas, &layerMap, "batch_norm",
|
||||
100, false, false);
|
||||
// test layer initialize
|
||||
std::vector<ParameterPtr> parameters;
|
||||
LayerPtr bnLayer;
|
||||
initTestLayer(configBN, &layerMap, ¶meters, &bnLayer);
|
||||
|
||||
std::vector<ParameterPtr> parameters2;
|
||||
LayerPtr convLayer;
|
||||
initTestLayer(config, &layerMap, ¶meters2, &convLayer);
|
||||
|
||||
bnLayer->forward(PASS_GC);
|
||||
convLayer->forward(PASS_GC);
|
||||
|
||||
CHECK_EQ(convLayer->getOutputValue()->getHeight(), 100);
|
||||
CHECK_EQ(convLayer->getOutputValue()->getWidth(), 576);
|
||||
}
|
||||
|
||||
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();
|
||||
}
|
||||
Loading…
Reference in new issue