def main():
args = parse_args()
gen = net.Generator1()
dis = net.Discriminator1()
clip_rect = None
if args.clip_rect:
clip_rect = map(int, args.clip_rect.split(','))
clip_rect = tuple([
clip_rect[0], clip_rect[1], clip_rect[0] + clip_rect[2],
clip_rect[1] + clip_rect[3]
])
gpu_device = None
if args.gpu >= 0:
device_id = args.gpu
cuda.get_device(device_id).use()
gen.to_gpu(device_id)
dis.to_gpu(device_id)
optimizer_gen = optimizers.Adam(alpha=0.001)
optimizer_gen.setup(gen)
optimizer_dis = optimizers.Adam(alpha=0.001)
optimizer_dis.setup(dis)
if args.input != None:
serializers.load_npz(args.input + '.gen.model', gen)
serializers.load_npz(args.input + '.gen.state', optimizer_gen)
serializers.load_npz(args.input + '.dis.model', dis)
serializers.load_npz(args.input + '.dis.state', optimizer_dis)
if args.out_image_dir != None:
if not os.path.exists(args.out_image_dir):
try:
os.mkdir(args.out_image_dir)
except:
print 'cannot make directory {}'.format(args.out_image_dir)
exit()
elif not os.path.isdir(args.out_image_dir):
print 'file path {} exists but is not directory'.format(
args.out_image_dir)
exit()
with open(args.dataset, 'rb') as f:
images = pickle.load(f)
train(gen,
dis,
optimizer_gen,
optimizer_dis,
images,
args.epoch,
batch_size=args.batch_size,
margin=args.margin,
save_epoch=args.save_epoch,
lr_decay=args.lr_decay,
output_path=args.output,
out_image_dir=args.out_image_dir,
clip_rect=clip_rect)
def main():
args = parse_args()
gen1 = net.Generator1()
gen2 = net.Generator2()
chainer.serializers.load_npz(args.model_path1, gen1)
chainer.serializers.load_npz(args.model_path2, gen2)
device_id = None
if args.gpu >= 0:
device_id = args.gpu
cuda.get_device(device_id).use()
gen1.to_gpu(device_id)
gen2.to_gpu(device_id)
out_vector_path = None
if args.vector_file1 and args.vector_index1 >= 0 and args.vector_file2 and args.vector_index2 >= 0:
with open(args.vector_file1, 'rb') as f:
z = np.load(f)
z1 = z[args.vector_index1]
with open(args.vector_file2, 'rb') as f:
z = np.load(f)
z2 = z[args.vector_index2]
w = np.arange(10).astype(np.float32).reshape((-1, 1)) / 9
z = (1 - w) * z1 + w * z2
z = z / (np.linalg.norm(z, axis=1, keepdims=True) + 1e-12)
else:
z = np.random.normal(0, 1, (100, latent_size)).astype(np.float32)
z = z / (np.linalg.norm(z, axis=1, keepdims=True) + 1e-12)
out_vector_path = '{}.npy'.format(args.output)
with chainer.no_backprop_mode():
if device_id is None:
x1 = gen1(z, train=False)
else:
x1 = gen1(cuda.to_gpu(z, device_id), train=False)
x2 = gen2(x1, train=False)
x1 = cuda.to_cpu(x1.data)
x2 = cuda.to_cpu(x2.data)
batch, ch, h, w = x1.shape
x1 = x1.reshape((-1, 10, ch, h, w)).transpose((0, 3, 1, 4, 2)).reshape(
(-1, 10 * w, ch))
x1 = ((x1 + 1) * 127.5).clip(0, 255).astype(np.uint8)
Image.fromarray(x1).save('{}_stack1.jpg'.format(args.output))
batch, ch, h, w = x2.shape
x2 = x2.reshape((-1, 10, ch, h, w)).transpose((0, 3, 1, 4, 2)).reshape(
(-1, 10 * w, ch))
x2 = ((x2 + 1) * 127.5).clip(0, 255).astype(np.uint8)
Image.fromarray(x2).save('{}_stack2.jpg'.format(args.output))
if out_vector_path:
with open(out_vector_path, 'wb') as f:
np.save(f, z)
import logging
import torchvision.utils as tov
from torch import Tensor as tensor
#pdb.set_trace()
gpuid = 0
lr_rate = 0.0002
alpha = 0.0005
beta = 0.01 * alpha
num_iter = 500000
optim_betas = (0.9, 0.999)
bs = 64
labelfake = Vb(torch.from_numpy(np.full((bs), 2, dtype=int))).cuda(gpuid)
labelpo = Vb(torch.from_numpy(np.full((bs), 1, dtype=int))).cuda(gpuid)
labelne = Vb(torch.from_numpy(np.full((bs), 0, dtype=int))).cuda(gpuid)
logging.basicConfig(filename='log/residualgan_v10.log', level=logging.INFO)
G1 = net.Generator1().cuda(gpuid)
G2 = net.Generator2().cuda(gpuid)
D = net.Discriminator().cuda(gpuid)
d_optimizer = optim.Adam(D.parameters(), lr=lr_rate, betas=optim_betas)
g1_optimizer = optim.Adam(G1.parameters(), lr=lr_rate, betas=optim_betas)
g2_optimizer = optim.Adam(G2.parameters(), lr=lr_rate, betas=optim_betas)
l1_crit = nn.L1Loss(size_average=False)
datalistpo = ld.getlist('../Eyeglasses_Positive.txt')
datalistne = ld.getlist('../Eyeglasses_Negative.txt')
iternow1 = 0
iternow2 = 0
for iter1 in xrange(num_iter):
D.zero_grad()
datapo, iternow1 = ld.load_data('../img_align_celeba_crop/',
'../Eyeglasses_Positive.txt',
datalistpo,