class GANModel:
def __init__(self, generator: Generator, loss: GANLoss, lr=0.0002):
self.generator = generator
self.generator.apply(weights_init)
self.loss = loss
self.loss.discriminator.apply(weights_init)
params = MinMaxParameters(self.generator.parameters(),
self.loss.parameters())
self.optimizer = MinMaxOptimizer(params, lr, lr * 2)
def discriminator_loss(self, real: Tensor, fake: Tensor) -> Loss:
return self.loss.discriminator_loss_with_penalty([real], [fake])
def generator_loss(self, real: Tensor, fake: Tensor) -> Loss:
return self.loss.generator_loss([real], [fake])
def loss_pair(self, real: Tensor, *noise: Tensor) -> MinMaxLoss:
fake = self.generator.forward(*noise)
return MinMaxLoss(self.generator_loss(real, fake),
self.discriminator_loss(real, fake))
def parameters(self) -> MinMaxParameters:
return MinMaxParameters(self.generator.parameters(),
self.loss.parameters())
def forward(self, real: Tensor, *noise: Tensor):
return self.loss_pair(real, *noise)
def train(self, real: Tensor, *noise: Tensor):
loss = self.loss_pair(real, *noise)
self.optimizer.train_step(loss)
return loss.min_loss.item(), loss.max_loss.item()
def __init__(self,
image_size: int,
generator_size: int = 32,
discriminator_size: int = 32,
channels_count: int = 3):
self.noise = NormalNoise(100, ParallelConfig.MAIN_DEVICE)
self.G = FillGenerator(self.noise, image_size, channels_count, channels_count, generator_size) \
.to(ParallelConfig.MAIN_DEVICE)
self.D = Discriminator(discriminator_size, 2 * channels_count, image_size) \
.to(ParallelConfig.MAIN_DEVICE)
self.G.apply(weights_init)
self.D.apply(weights_init)
if ParallelConfig.GPU_IDS.__len__() > 1:
self.G = nn.DataParallel(self.G, ParallelConfig.GPU_IDS)
self.D = nn.DataParallel(self.D, ParallelConfig.GPU_IDS)
was_loss = WassersteinLoss(2) \
.add_penalty(AdaptiveLipschitzPenalty(0.1, 0.01)) \
.add_penalty(L2Penalty(0.1))
self.gan_model = ConditionalGANModel(self.G, self.D, was_loss)
lr = 0.0002
self.optimizer = MinMaxOptimizer(self.gan_model.parameters(), lr, lr)
def __init__(self,
image_size: int,
mask_channels_count: int,
image_channels_count: int = 3,
noise=NormalNoise(50, ParallelConfig.MAIN_DEVICE),
generator_size: int = 32,
discriminator_size: int = 32):
netG = UNetGenerator(noise, image_size, mask_channels_count, image_channels_count, generator_size) \
.to(ParallelConfig.MAIN_DEVICE)
netD = Discriminator(discriminator_size, image_channels_count + mask_channels_count, image_size) \
.to(ParallelConfig.MAIN_DEVICE)
netG.apply(weights_init)
netD.apply(weights_init)
if torch.cuda.device_count() > 1:
netD = nn.DataParallel(netD, ParallelConfig.GPU_IDS)
netG = nn.DataParallel(netG, ParallelConfig.GPU_IDS)
self.gan_model = ConditionalGANModel(
netG, netD,
WassersteinLoss(10.0)
# .add_penalty(AdaptiveLipschitzPenalty(1, 0.05))
# .add_penalty(L2Penalty(0.01))
)
lrG = 0.0001
lrD = 0.0004
self.optimizer = MinMaxOptimizer(self.gan_model.parameters(), lrG, lrD)
def __init__(self, generator: Generator, loss: GANLoss, lr=0.0002):
self.generator = generator
self.generator.apply(weights_init)
self.loss = loss
self.loss.discriminator.apply(weights_init)
params = MinMaxParameters(self.generator.parameters(),
self.loss.parameters())
self.optimizer = MinMaxOptimizer(params, lr, lr * 2)
def __init__(self,
generator: ConditionalGenerator,
loss: GANLoss,
lr=0.0002):
self.generator = generator
self.loss = loss
params = MinMaxParameters(self.generator.parameters(),
self.loss.parameters())
self.optimizer = MinMaxOptimizer(params, lr, lr * 2)
class MaskToImageComposite:
def __init__(self,
image_size: int,
labels_list: List[int],
image_channels_count: int = 3,
noise=NormalNoise(100, ParallelConfig.MAIN_DEVICE),
generator_size: int = 32,
discriminator_size: int = 32):
mask_nc = len(labels_list)
gen_list = nn.ModuleList([
UNetGenerator(noise,
image_size,
1,
image_channels_count,
int(generator_size / 2),
nc_max=256) for i in range(mask_nc)
])
netG = CompositeGenerator(noise, gen_list) \
.to(ParallelConfig.MAIN_DEVICE)
netD = Discriminator(discriminator_size, image_channels_count + mask_nc, image_size) \
.to(ParallelConfig.MAIN_DEVICE)
netG.apply(weights_init)
netD.apply(weights_init)
if torch.cuda.device_count() > 1:
netD = nn.DataParallel(netD, ParallelConfig.GPU_IDS)
netG = nn.DataParallel(netG, ParallelConfig.GPU_IDS)
self.gan_model = ConditionalGANModel(
netG, netD,
WassersteinLoss(2).add_penalty(AdaptiveLipschitzPenalty(
0.1, 0.01)).add_penalty(L2Penalty(0.1)) +
VggGeneratorLoss(15, 1))
# vgg_loss_fn = VggGeneratorLoss(ParallelConfig.MAIN_DEVICE)
lrG = 0.0002
lrD = 0.0002
self.optimizer = MinMaxOptimizer(self.gan_model.parameters(), lrG, lrD)
def train(self, images: Tensor, masks: Mask):
loss: MinMaxLoss = self.gan_model.loss_pair(images, masks.tensor)
self.optimizer.train_step(loss)
def generator_loss(self, images: Tensor, masks: Mask) -> Loss:
fake = self.gan_model.generator.forward(masks.tensor)
return self.gan_model.generator_loss(images, fake, masks.tensor)
class FillImageModel:
def __init__(self,
image_size: int,
generator_size: int = 32,
discriminator_size: int = 32,
channels_count: int = 3):
self.noise = NormalNoise(100, ParallelConfig.MAIN_DEVICE)
self.G = FillGenerator(self.noise, image_size, channels_count, channels_count, generator_size) \
.to(ParallelConfig.MAIN_DEVICE)
self.D = Discriminator(discriminator_size, 2 * channels_count, image_size) \
.to(ParallelConfig.MAIN_DEVICE)
self.G.apply(weights_init)
self.D.apply(weights_init)
if ParallelConfig.GPU_IDS.__len__() > 1:
self.G = nn.DataParallel(self.G, ParallelConfig.GPU_IDS)
self.D = nn.DataParallel(self.D, ParallelConfig.GPU_IDS)
was_loss = WassersteinLoss(2) \
.add_penalty(AdaptiveLipschitzPenalty(0.1, 0.01)) \
.add_penalty(L2Penalty(0.1))
self.gan_model = ConditionalGANModel(self.G, self.D, was_loss)
lr = 0.0002
self.optimizer = MinMaxOptimizer(self.gan_model.parameters(), lr, lr)
def train(self, images: Tensor, segments: Mask):
front: Tensor = images * segments.tensor
loss: MinMaxLoss = self.gan_model.loss_pair(images, front,
segments.tensor)
self.optimizer.train_step(loss)
def test(self, images: Tensor, segments: Mask) -> Tensor:
front: Tensor = images * segments.tensor
return self.G(front, segments.tensor)
def generator_loss(self, images: Tensor, segments: Mask) -> Loss:
front: Tensor = images * segments.tensor
fake = self.G(front, segments.tensor)
loss = self.gan_model.generator_loss(images, fake, front)
return loss
class MaskToImage:
def __init__(self,
image_size: int,
mask_channels_count: int,
image_channels_count: int = 3,
noise=NormalNoise(50, ParallelConfig.MAIN_DEVICE),
generator_size: int = 32,
discriminator_size: int = 32):
netG = UNetGenerator(noise, image_size, mask_channels_count, image_channels_count, generator_size) \
.to(ParallelConfig.MAIN_DEVICE)
netD = Discriminator(discriminator_size, image_channels_count + mask_channels_count, image_size) \
.to(ParallelConfig.MAIN_DEVICE)
netG.apply(weights_init)
netD.apply(weights_init)
if torch.cuda.device_count() > 1:
netD = nn.DataParallel(netD, ParallelConfig.GPU_IDS)
netG = nn.DataParallel(netG, ParallelConfig.GPU_IDS)
self.gan_model = ConditionalGANModel(
netG, netD,
WassersteinLoss(10.0)
# .add_penalty(AdaptiveLipschitzPenalty(1, 0.05))
# .add_penalty(L2Penalty(0.01))
)
lrG = 0.0001
lrD = 0.0004
self.optimizer = MinMaxOptimizer(self.gan_model.parameters(), lrG, lrD)
def train(self, images: Tensor, masks: Mask):
loss: MinMaxLoss = self.gan_model.loss_pair(images, masks.tensor)
self.optimizer.train_step(loss)
def generator_loss(self, images: Tensor, masks: Mask) -> Loss:
fake = self.gan_model.generator.forward(masks.tensor)
return self.gan_model.generator_loss(images, fake, masks.tensor)
device = torch.device("cuda:1")
noise = NormalNoise(noise_size, device)
netG = EGenerator(noise).to(device)
print(netG)
netD = EDiscriminator().to(device)
print(netD)
lr = 0.001
betas = (0.5, 0.999)
gan_model = GANModel(netG, netD, HingeLoss())
optimizer = MinMaxOptimizer(gan_model.parameters(), lr, 2 * lr, betas)
n = 5000
xs = (torch.arange(0, n, dtype=torch.float32) / 100.0).view(n, 1)
ys = torch.cat((xs.cos(), xs.sin()), dim=1)
plt.scatter(ys[:, 0].view(n).numpy(), ys[:, 1].view(n).numpy())
print("Starting Training Loop...")
def gen_batch() -> Tensor:
i = random.randint(0, n - batch_size)
j = i + batch_size
return ys[i:j, :]