mindspore/tests/st/probability/test_gpu_vae_gan.py

# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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.
# ============================================================================
"""
The VAE interface can be called to construct VAE-GAN network.
"""
import os

import mindspore.dataset as ds
import mindspore.dataset.vision.c_transforms as CV
import mindspore.nn as nn
from mindspore import context
from mindspore.ops import operations as P
from mindspore.ops import composite as C
from mindspore.nn.probability.dpn import VAE
from mindspore.nn.probability.infer import ELBO, SVI

context.set_context(mode=context.GRAPH_MODE, save_graphs=False, device_target="GPU")
IMAGE_SHAPE = (-1, 1, 32, 32)
image_path = os.path.join('/home/workspace/mindspore_dataset/mnist', "train")


class Encoder(nn.Cell):
    def __init__(self):
        super(Encoder, self).__init__()
        self.fc1 = nn.Dense(1024, 400)
        self.relu = nn.ReLU()
        self.flatten = nn.Flatten()

    def construct(self, x):
        x = self.flatten(x)
        x = self.fc1(x)
        x = self.relu(x)
        return x


class Decoder(nn.Cell):
    def __init__(self):
        super(Decoder, self).__init__()
        self.fc1 = nn.Dense(400, 1024)
        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        self.reshape = P.Reshape()

    def construct(self, z):
        z = self.fc1(z)
        z = self.reshape(z, IMAGE_SHAPE)
        z = self.sigmoid(z)
        return z


class Discriminator(nn.Cell):
    """
    The Discriminator of the GAN network.
    """

    def __init__(self):
        super(Discriminator, self).__init__()
        self.fc1 = nn.Dense(1024, 400)
        self.fc2 = nn.Dense(400, 720)
        self.fc3 = nn.Dense(720, 1024)
        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        self.flatten = nn.Flatten()

    def construct(self, x):
        x = self.flatten(x)
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        x = self.relu(x)
        x = self.fc3(x)
        x = self.sigmoid(x)
        return x


class VaeGan(nn.Cell):
    def __init__(self):
        super(VaeGan, self).__init__()
        self.E = Encoder()
        self.G = Decoder()
        self.D = Discriminator()
        self.dense = nn.Dense(20, 400)
        self.vae = VAE(self.E, self.G, 400, 20)
        self.shape = P.Shape()
        self.normal = C.normal
        self.to_tensor = P.ScalarToArray()

    def construct(self, x):
        recon_x, x, mu, std = self.vae(x)
        z_p = self.normal(self.shape(mu), self.to_tensor(0.0), self.to_tensor(1.0), seed=0)
        z_p = self.dense(z_p)
        x_p = self.G(z_p)
        ld_real = self.D(x)
        ld_fake = self.D(recon_x)
        ld_p = self.D(x_p)
        return ld_real, ld_fake, ld_p, recon_x, x, mu, std


class VaeGanLoss(ELBO):
    def __init__(self):
        super(VaeGanLoss, self).__init__()
        self.zeros = P.ZerosLike()
        self.mse = nn.MSELoss(reduction='sum')

    def construct(self, data, label):
        ld_real, ld_fake, ld_p, recon_x, x, mu, std = data
        y_real = self.zeros(ld_real) + 1
        y_fake = self.zeros(ld_fake)
        loss_D = self.mse(ld_real, y_real)
        loss_GD = self.mse(ld_p, y_fake)
        loss_G = self.mse(ld_fake, y_real)
        reconstruct_loss = self.recon_loss(x, recon_x)
        kl_loss = self.posterior('kl_loss', 'Normal', self.zeros(mu), self.zeros(mu) + 1, mu, std)
        elbo_loss = reconstruct_loss + self.sum(kl_loss)
        return loss_D + loss_G + loss_GD + elbo_loss


def create_dataset(data_path, batch_size=32, repeat_size=1,
                   num_parallel_workers=1):
    """
    create dataset for train or test
    """
    # define dataset
    mnist_ds = ds.MnistDataset(data_path)

    resize_height, resize_width = 32, 32
    rescale = 1.0 / 255.0
    shift = 0.0

    # define map operations
    resize_op = CV.Resize((resize_height, resize_width))  # Bilinear mode
    rescale_op = CV.Rescale(rescale, shift)
    hwc2chw_op = CV.HWC2CHW()

    # apply map operations on images
    mnist_ds = mnist_ds.map(operations=resize_op, input_columns="image", num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(operations=rescale_op, input_columns="image", num_parallel_workers=num_parallel_workers)
    mnist_ds = mnist_ds.map(operations=hwc2chw_op, input_columns="image", num_parallel_workers=num_parallel_workers)

    # apply DatasetOps
    mnist_ds = mnist_ds.batch(batch_size)
    mnist_ds = mnist_ds.repeat(repeat_size)

    return mnist_ds


def test_vae_gan():
    vae_gan = VaeGan()
    net_loss = VaeGanLoss()
    optimizer = nn.Adam(params=vae_gan.trainable_params(), learning_rate=0.001)
    ds_train = create_dataset(image_path, 128, 1)
    net_with_loss = nn.WithLossCell(vae_gan, net_loss)
    vi = SVI(net_with_loss=net_with_loss, optimizer=optimizer)
    vae_gan = vi.run(train_dataset=ds_train, epochs=5)