原文链接：https://arxiv.org/pdf/1604.07379.pdf
简介
背景：从“Conditinal GAN”到“InfoGAN”，我们都在致力于解决一个问题就是如何通过人为控制的范式，来控制GAN网络生成我们想要的数据，但是之前的几种方法大多是在针对模型的输入与输出在做文章。而本文是通过构思了一种双（或多）GAN的架构，通过不同域的相互约束来达到控制生成数据的样式。
核心思想：使用一对GAN通过权重共享的方式，是的两个数据空间域同时约束生成数据。
由上图来看，我们从中间划开来看的话就是两个独立的GAN，但是有所不同的是这两个GAN在生成器前几层与生成器后几层是共享像网络权重的。为什么我们要共享网络权重呢？可以感性地这样理解，我们将GAN网络应用到了两个任务中，于是乎GAN会受到两个任务的约束，而约束越多就越方便我们控制GAN的优化方向。
基础结构
生成器
我们将生成器的每一层网络拆分开，即可得到上面的式子，由于两个生成器是从相同的随机变量映射到不同的数据空间，因此共享的网络层只能是接近噪声z输入端的，即m较小的层。
判别器
我们将判别器的每一层网络拆分开，即可得到上面的式子，由于两个判别器是从不同的数据空间映射到真假的判断结果，因此共享的网络层只能是接近结果输出端的，即n较大的层。
LOSS
用一张简单的示意图表示：
上图两个大的矩形就是联合分布域形成的约束。
代码与实践结果
参考链接：https://github.com/WingsofFAN/PyTorch-GAN/blob/master/implementations/cogan/cogan.py
import argparse
import os
import numpy as np
import math
import scipy
import itertools
import mnistm
import torchvision.transforms as transforms
from torchvision.utils import save_image
from torch.utils.data import DataLoader
from torchvision import datasets
from torch.autograd import Variable
import torch.nn as nn
import torch.nn.functional as F
import torch
os.makedirs("images", exist_ok=True)
parser = argparse.ArgumentParser()
parser.add_argument("--n_epochs", type=int, default=200, help="number of epochs of training")
parser.add_argument("--batch_size", type=int, default=32, help="size of the batches")
parser.add_argument("--lr", type=float, default=0.0002, help="adam: learning rate")
parser.add_argument("--b1", type=float, default=0.5, help="adam: decay of first order momentum of gradient")
parser.add_argument("--b2", type=float, default=0.999, help="adam: decay of first order momentum of gradient")
parser.add_argument("--n_cpu", type=int, default=8, help="number of cpu threads to use during batch generation")
parser.add_argument("--latent_dim", type=int, default=100, help="dimensionality of the latent space")
parser.add_argument("--img_size", type=int, default=32, help="size of each image dimension")
parser.add_argument("--channels", type=int, default=3, help="number of image channels")
parser.add_argument("--sample_interval", type=int, default=400, help="interval betwen image samples")
opt = parser.parse_args()
print(opt)
img_shape = (opt.channels, opt.img_size, opt.img_size)
cuda = True if torch.cuda.is_available() else False
def weights_init_normal(m):
classname = m.__class__.__name__
if classname.find("Linear") != -1:
torch.nn.init.normal_(m.weight.data, 0.0, 0.02)
elif classname.find("BatchNorm") != -1:
torch.nn.init.normal_(m.weight.data, 1.0, 0.02)
torch.nn.init.constant_(m.bias.data, 0.0)
class CoupledGenerators(nn.Module):
def __init__(self):
super(CoupledGenerators, self).__init__()
self.init_size = opt.img_size // 4
self.fc = nn.Sequential(nn.Linear(opt.latent_dim, 128 * self.init_size ** 2))
self.shared_conv = nn.Sequential(
nn.BatchNorm2d(128),
nn.Upsample(scale_factor=2),
nn.Conv2d(128, 128, 3, stride=1, padding=1),
nn.BatchNorm2d(128, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Upsample(scale_factor=2),
)
self.G1 = nn.Sequential(
nn.Conv2d(128, 64, 3, stride=1, padding=1),
nn.BatchNorm2d(64, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(64, opt.channels, 3, stride=1, padding=1),
nn.Tanh(),
)
self.G2 = nn.Sequential(
nn.Conv2d(128, 64, 3, stride=1, padding=1),
nn.BatchNorm2d(64, 0.8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(64, opt.channels, 3, stride=1, padding=1),
nn.Tanh(),
)
def forward(self, noise):
out = self.fc(noise)
out = out.view(out.shape[0], 128, self.init_size, self.init_size)
img_emb = self.shared_conv(out)
img1 = self.G1(img_emb)
img2 = self.G2(img_emb)
return img1, img2
class CoupledDiscriminators(nn.Module):
def __init__(self):
super(CoupledDiscriminators, self).__init__()
def discriminator_block(in_filters, out_filters, bn=True):
block = [nn.Conv2d(in_filters, out_filters, 3, 2, 1)]
if bn:
block.append(nn.BatchNorm2d(out_filters, 0.8))
block.extend([nn.LeakyReLU(0.2, inplace=True), nn.Dropout2d(0.25)])
return block
self.shared_conv = nn.Sequential(
*discriminator_block(opt.channels, 16, bn=False),
*discriminator_block(16, 32),
*discriminator_block(32, 64),
*discriminator_block(64, 128),
)
# The height and width of downsampled image
ds_size = opt.img_size // 2 ** 4
self.D1 = nn.Linear(128 * ds_size ** 2, 1)
self.D2 = nn.Linear(128 * ds_size ** 2, 1)
def forward(self, img1, img2):
# Determine validity of first image
out = self.shared_conv(img1)
out = out.view(out.shape[0], -1)
validity1 = self.D1(out)
# Determine validity of second image
out = self.shared_conv(img2)
out = out.view(out.shape[0], -1)
validity2 = self.D2(out)
return validity1, validity2
# Loss function
adversarial_loss = torch.nn.MSELoss()
# Initialize models
coupled_generators = CoupledGenerators()
coupled_discriminators = CoupledDiscriminators()
if cuda:
coupled_generators.cuda()
coupled_discriminators.cuda()
# Initialize weights
coupled_generators.apply(weights_init_normal)
coupled_discriminators.apply(weights_init_normal)
# Configure data loader
os.makedirs("../../data/mnist", exist_ok=True)
dataloader1 = torch.utils.data.DataLoader(
datasets.MNIST(
"../../data/mnist",
train=True,
download=True,
transform=transforms.Compose(
[transforms.Resize(opt.img_size), transforms.ToTensor(), transforms.Normalize([0.5], [0.5])]
),
),
batch_size=opt.batch_size,
shuffle=True,
)
os.makedirs("../../data/mnistm", exist_ok=True)
dataloader2 = torch.utils.data.DataLoader(
mnistm.MNISTM(
"../../data/mnistm",
train=True,
download=True,
transform=transforms.Compose(
[
transforms.Resize(opt.img_size),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
]
),
),
batch_size=opt.batch_size,
shuffle=True,
)
# Optimizers
optimizer_G = torch.optim.Adam(coupled_generators.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))
optimizer_D = torch.optim.Adam(coupled_discriminators.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))
Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
# ----------
# Training
# ----------
for epoch in range(opt.n_epochs):
for i, ((imgs1, _), (imgs2, _)) in enumerate(zip(dataloader1, dataloader2)):
batch_size = imgs1.shape[0]
# Adversarial ground truths
valid = Variable(Tensor(batch_size, 1).fill_(1.0), requires_grad=False)
fake = Variable(Tensor(batch_size, 1).fill_(0.0), requires_grad=False)
# Configure input
imgs1 = Variable(imgs1.type(Tensor).expand(imgs1.size(0), 3, opt.img_size, opt.img_size))
imgs2 = Variable(imgs2.type(Tensor))
# ------------------
# Train Generators
# ------------------
optimizer_G.zero_grad()
# Sample noise as generator input
z = Variable(Tensor(np.random.normal(0, 1, (batch_size, opt.latent_dim))))
# Generate a batch of images
gen_imgs1, gen_imgs2 = coupled_generators(z)
# Determine validity of generated images
validity1, validity2 = coupled_discriminators(gen_imgs1, gen_imgs2)
g_loss = (adversarial_loss(validity1, valid) + adversarial_loss(validity2, valid)) / 2
g_loss.backward()
optimizer_G.step()
# ----------------------
# Train Discriminators
# ----------------------
optimizer_D.zero_grad()
# Determine validity of real and generated images
validity1_real, validity2_real = coupled_discriminators(imgs1, imgs2)
validity1_fake, validity2_fake = coupled_discriminators(gen_imgs1.detach(), gen_imgs2.detach())
#真实图片输入对应两个loss
#生成图片输入对应两个loss
d_loss = (
adversarial_loss(validity1_real, valid)
+ adversarial_loss(validity1_fake, fake)
+ adversarial_loss(validity2_real, valid)
+ adversarial_loss(validity2_fake, fake)
) / 4
d_loss.backward()
optimizer_D.step()
print(
"[Epoch %d/%d] [Batch %d/%d] [D loss: %f] [G loss: %f]"
% (epoch, opt.n_epochs, i, len(dataloader1), d_loss.item(), g_loss.item())
)
batches_done = epoch * len(dataloader1) + i
if batches_done % opt.sample_interval == 0:
gen_imgs = torch.cat((gen_imgs1.data, gen_imgs2.data), 0)
save_image(gen_imgs, "images/%d.png" % batches_done, nrow=8, normalize=True)
mnist与mnistm测试结果
mnist mnistm