class Experiment009(Experiment001):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Decoder and Generator shares parameters.
- Update Generator0 only when training generator, i.e., not train Decoder.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_001 \
import Encoder, Decoder, Generator0, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator0)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def generate(self, x_l, test):
xp = cuda.get_array_module(x_l)
z = Variable(
cuda.to_gpu(
xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z, test), test)
return x_gen
def cleargrads(self, ):
self.encoder.cleargrads()
self.decoder.cleargrads()
self.generator0.cleargrads()
self.discriminator.cleargrads()
def serialize(self, epoch, filename):
# Create dir path
dpath = os.path.join(filename, "./model_{:05d}".format(epoch))
if os.path.exists(dpath):
shutil.rmtree(dpath)
os.makedirs(dpath)
else:
os.makedirs(dpath)
# Serialize
fpath = os.path.join(dpath, "encoder.h5py")
serializers.save_hdf5(fpath, self.encoder)
fpath = os.path.join(dpath, "decoder.h5py")
serializers.save_hdf5(fpath, self.decoder)
fpath = os.path.join(dpath, "generator0.h5py")
serializers.save_hdf5(fpath, self.generator0)
class Experiment006(Experiment005):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Discriminator is conditioned on hidden feature of x_real.
- Decoder and Generator shares parameters.
- Decoder and Generator shares optimizer.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_003 \
import Encoder, Decoder, Discriminator, Generator0
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.generator = self.decoder
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator0)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
hz = self.generator0(z)
x_rec = self.decoder(h, hz)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
h.unchain_backward()
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
hz = self.generator0(z)
x_gen = self.generator(h, hz)
d_x_gen = self.discriminator(x_gen, h)
d_x_real = self.discriminator(x, h)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
hz = self.generator0(z)
x_gen = self.generator(h, hz)
d_x_gen = self.discriminator(x_gen, h)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_dec.update()
self.optimizer_gen.update()
class Experiment008(Experiment003):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Discriminator is conditioned on hidden feature of x_real.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_002 \
import Encoder, Decoder, Generator, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator = Generator(device, act, dim)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen, h)
d_x_real = self.discriminator(x, h)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen, h)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def test(self, x_l, y_l, epoch, filename):
"""generate samples, then save"""
x_gen = self.generate(x_l, test=True)
self.save(x_l, x_gen, epoch, filename)
h = self.encoder(x_l, test=True)
d_x_gen = self.discriminator(x_gen, h, test=True)
loss = self.lsgan_loss(d_x_gen)
return loss
class Experiment000(object):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Feature matching is taken between convolution ouputs.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_000 \
import Encoder, Decoder, Generator, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator = Generator(device, act, dim)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen) + self.recon_loss(h, h_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def test(self, x_l, y_l, epoch, filename):
"""generate samples, then save"""
x_gen = self.generate(x_l, test=True)
self.save(x_l, x_gen, epoch, filename)
d_x_gen = self.discriminator(x_gen, test=True)
loss = self.lsgan_loss(d_x_gen)
return loss
def generate(self, x_l, test):
h = self.encoder(x_l, test)
xp = cuda.get_array_module(x_l)
z = Variable(
cuda.to_gpu(
xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.generator(h, z, test)
return x_gen
def save(self, x_l, x_gen, epoch, filename):
# Create dir path for genenrated images
dpath_gen = os.path.join(filename, "gen",
"images_{:05d}".format(epoch))
if os.path.exists(dpath_gen):
shutil.rmtree(dpath_gen)
os.makedirs(dpath_gen)
else:
os.makedirs(dpath_gen)
# Create dir path for real images
dpath_real = os.path.join(filename, "real",
"images_{:05d}".format(epoch))
if os.path.exists(dpath_real):
shutil.rmtree(dpath_real)
os.makedirs(dpath_real)
else:
os.makedirs(dpath_real)
# Save labeled images
imgs = cuda.to_cpu(x_l.data)
imgs = 127.5 * imgs + 127.5
for i, img in enumerate(imgs):
fpath = os.path.join(dpath_real, "{:05d}.png".format(i))
cv2.imwrite(fpath, np.squeeze(img))
# Save generated images
imgs = cuda.to_cpu(x_gen.data)
imgs = 127.5 * imgs + 127.5
for i, img in enumerate(imgs):
fpath = os.path.join(dpath_gen, "{:05d}.png".format(i))
cv2.imwrite(fpath, np.squeeze(img))
def serialize(self, epoch, filename):
# Create dir path
dpath = os.path.join(filename, "./model_{:05d}".format(epoch))
if os.path.exists(dpath):
shutil.rmtree(dpath)
os.makedirs(dpath)
else:
os.makedirs(dpath)
# Serialize
fpath = os.path.join(dpath, "encoder.h5py")
serializers.save_hdf5(fpath, self.encoder)
fpath = os.path.join(dpath, "generator.h5py")
serializers.save_hdf5(fpath, self.generator)
def cleargrads(self, ):
self.encoder.cleargrads()
self.decoder.cleargrads()
self.generator.cleargrads()
self.discriminator.cleargrads()
class Experiment002(Experiment001):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Feature matching is taken between convolution ouputs.
- Decoder and Generator shares parameters.
- Update Generator0 and Decoder when training generator.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_001 \
import Encoder, Decoder, Generator0, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen0 = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen0.setup(self.generator0)
self.optimizer_gen0.use_cleargrads()
self.optimizer_gen1 = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen1.setup(self.decoder)
self.optimizer_gen1.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(
cuda.to_gpu(
xp.random.rand(x.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen) + self.recon_loss(h, h_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen0.update()
self.optimizer_gen1.update()
def generate(self, x_l, test):
xp = cuda.get_array_module(x_l)
z = Variable(
cuda.to_gpu(
xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32),
self.device))
x_gen = self.decoder(self.generator0(z, test), test)
return x_gen
class Experiment010(Experiment002):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Decoder and Generator shares parameters.
- Update Generator0 and Decoder when training generator.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_001 \
import Encoder, Decoder, Generator0, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen0 = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen0.setup(self.generator0)
self.optimizer_gen0.use_cleargrads()
self.optimizer_gen1 = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen1.setup(self.decoder)
self.optimizer_gen1.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen0.update()
self.optimizer_gen1.update()
def generate(self, x_l, test):
xp = cuda.get_array_module(x_l)
z = Variable(cuda.to_gpu(xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z, test), test)
return x_gen
class Experiment009(Experiment001):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Decoder and Generator shares parameters.
- Update Generator0 only when training generator, i.e., not train Decoder.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_001 \
import Encoder, Decoder, Generator0, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator0)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z))
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def generate(self, x_l, test):
xp = cuda.get_array_module(x_l)
z = Variable(cuda.to_gpu(xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.decoder(self.generator0(z, test), test)
return x_gen
def cleargrads(self, ):
self.encoder.cleargrads()
self.decoder.cleargrads()
self.generator0.cleargrads()
self.discriminator.cleargrads()
def serialize(self, epoch, filename):
# Create dir path
dpath = os.path.join(filename, "./model_{:05d}".format(epoch))
if os.path.exists(dpath):
shutil.rmtree(dpath)
os.makedirs(dpath)
else:
os.makedirs(dpath)
# Serialize
fpath = os.path.join(dpath, "encoder.h5py")
serializers.save_hdf5(fpath, self.encoder)
fpath = os.path.join(dpath, "decoder.h5py")
serializers.save_hdf5(fpath, self.decoder)
fpath = os.path.join(dpath, "generator0.h5py")
serializers.save_hdf5(fpath, self.generator0)
class Experiment008(Experiment003):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Discriminator is conditioned on hidden feature of x_real.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_002 \
import Encoder, Decoder, Generator, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator = Generator(device, act ,dim)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen, h)
d_x_real = self.discriminator(x, h)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen, h)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def test(self, x_l, y_l, epoch, filename):
"""generate samples, then save"""
x_gen = self.generate(x_l, test=True)
self.save(x_l, x_gen, epoch, filename)
h = self.encoder(x_l, test=True)
d_x_gen = self.discriminator(x_gen, h, test=True)
loss = self.lsgan_loss(d_x_gen)
return loss
class Experiment006(Experiment005):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Discriminator is conditioned on hidden feature of x_real.
- Decoder and Generator shares parameters.
- Decoder and Generator shares optimizer.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_003 \
import Encoder, Decoder, Discriminator, Generator0
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator0 = Generator0(dim, device, act)
self.generator = self.decoder
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator0.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator0)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
hz = self.generator0(z)
x_rec = self.decoder(h, hz)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
h.unchain_backward()
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
hz = self.generator0(z)
x_gen = self.generator(h, hz)
d_x_gen = self.discriminator(x_gen, h)
d_x_real = self.discriminator(x, h)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
hz = self.generator0(z)
x_gen = self.generator(h, hz)
d_x_gen = self.discriminator(x_gen, h)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_dec.update()
self.optimizer_gen.update()
class Experiment000(object):
"""Enc-Dec, Enc-Gen-Enc, Enc-Gen-Dis.
- Feature matching is taken between convolution ouputs.
"""
def __init__(self, device=None, learning_rate=1e-3, act=F.relu, dim=100):
# Settings
self.device = device
self.act = act
self.learning_rate = learning_rate
self.dim = dim
# Losses
self.recon_loss = ReconstructionLoss()
self.lsgan_loss = LSGANLoss()
# Model
from sslgen2.mnist.cnn_model_000 \
import Encoder, Decoder, Generator, Discriminator
self.encoder = Encoder(device, act)
self.decoder = Decoder(device, act)
self.generator = Generator(device, act ,dim)
self.discriminator = Discriminator(device, act)
self.encoder.to_gpu(device) if self.device else None
self.decoder.to_gpu(device) if self.device else None
self.generator.to_gpu(device) if self.device else None
self.discriminator.to_gpu(device) if self.device else None
# Optimizer
self.optimizer_enc = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_enc.setup(self.encoder)
self.optimizer_enc.use_cleargrads()
self.optimizer_dec = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dec.setup(self.decoder)
self.optimizer_dec.use_cleargrads()
self.optimizer_gen = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_gen.setup(self.generator)
self.optimizer_gen.use_cleargrads()
self.optimizer_dis = optimizers.Adam(learning_rate, beta1=0.5)
self.optimizer_dis.setup(self.discriminator)
self.optimizer_dis.use_cleargrads()
def train(self, x):
# Encoder/Decoder
h = self.encoder(x)
x_rec = self.decoder(h)
l_rec = self.recon_loss(x, x_rec)
self.cleargrads()
l_rec.backward()
self.optimizer_enc.update()
self.optimizer_dec.update()
# Discriminator
h = Variable(h.data) # disconnect
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen)
d_x_real = self.discriminator(x)
l_dis = self.lsgan_loss(d_x_gen, d_x_real)
self.cleargrads()
l_dis.backward()
self.optimizer_dis.update()
# Generator
xp = cuda.get_array_module(x)
z = Variable(cuda.to_gpu(xp.random.rand(x.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.generator(h, z)
d_x_gen = self.discriminator(x_gen)
h_gen = self.encoder(x_gen)
l_gen = self.lsgan_loss(d_x_gen) + self.recon_loss(h, h_gen)
self.cleargrads()
l_gen.backward()
self.optimizer_gen.update()
def test(self, x_l, y_l, epoch, filename):
"""generate samples, then save"""
x_gen = self.generate(x_l, test=True)
self.save(x_l, x_gen, epoch, filename)
d_x_gen = self.discriminator(x_gen, test=True)
loss = self.lsgan_loss(d_x_gen)
return loss
def generate(self, x_l, test):
h = self.encoder(x_l, test)
xp = cuda.get_array_module(x_l)
z = Variable(cuda.to_gpu(xp.random.rand(x_l.shape[0], self.dim).astype(xp.float32), self.device))
x_gen = self.generator(h, z, test)
return x_gen
def save(self, x_l, x_gen, epoch, filename):
# Create dir path for genenrated images
dpath_gen = os.path.join(filename,
"gen",
"images_{:05d}".format(epoch))
if os.path.exists(dpath_gen):
shutil.rmtree(dpath_gen)
os.makedirs(dpath_gen)
else:
os.makedirs(dpath_gen)
# Create dir path for real images
dpath_real = os.path.join(filename,
"real",
"images_{:05d}".format(epoch))
if os.path.exists(dpath_real):
shutil.rmtree(dpath_real)
os.makedirs(dpath_real)
else:
os.makedirs(dpath_real)
# Save labeled images
imgs = cuda.to_cpu(x_l.data)
imgs = 127.5 * imgs + 127.5
for i, img in enumerate(imgs):
fpath = os.path.join(dpath_real, "{:05d}.png".format(i))
cv2.imwrite(fpath, np.squeeze(img))
# Save generated images
imgs = cuda.to_cpu(x_gen.data)
imgs = 127.5 * imgs + 127.5
for i, img in enumerate(imgs):
fpath = os.path.join(dpath_gen, "{:05d}.png".format(i))
cv2.imwrite(fpath, np.squeeze(img))
def serialize(self, epoch, filename):
# Create dir path
dpath = os.path.join(filename, "./model_{:05d}".format(epoch))
if os.path.exists(dpath):
shutil.rmtree(dpath)
os.makedirs(dpath)
else:
os.makedirs(dpath)
# Serialize
fpath = os.path.join(dpath, "encoder.h5py")
serializers.save_hdf5(fpath, self.encoder)
fpath = os.path.join(dpath, "generator.h5py")
serializers.save_hdf5(fpath, self.generator)
def cleargrads(self, ):
self.encoder.cleargrads()
self.decoder.cleargrads()
self.generator.cleargrads()
self.discriminator.cleargrads()