for i in range(args.n_attrs):
tmp = att_a.clone()
for j in range(args.n_attrs):
tmp[:, j] = -1
tmp[:, i] = 1
tmp = check_attribute_conflict(tmp, args.attrs[i], args.attrs)
att_b_list.append(tmp)
else:
for i in range(args.n_attrs):
tmp = att_a.clone()
tmp[:, i] = 1 - tmp[:, i]
tmp = check_attribute_conflict(tmp, args.attrs[i], args.attrs)
att_b_list.append(tmp) # a list of lists of attributes (probably).
with torch.no_grad():
samples = [img_a]
for i, att_b in enumerate(att_b_list):
att_b_ = (att_b * 2 - 1) * args.thres_int
if i > 0:
att_b_[..., i - 1] = att_b_[..., i - 1] * args.test_int / args.thres_int
samples.append(attgan.G(img_a, att_b_))
samples = torch.cat(samples, dim=3)
if args.custom_img:
out_file = test_dataset.images[idx]
else:
out_file = '{:06d}.jpg'.format(idx + 182638)
vutils.save_image(
samples, join(output_path, out_file),
nrow=1, normalize=True, range=(-1., 1.)
)
print('{:s} done!'.format(out_file))
os.path.join('output', args.experiment_name, 'checkpoint',
'weights.{:d}.pth'.format(epoch)))
# attgan.save(os.path.join(
# 'output', args.experiment_name, 'checkpoint', 'weights.{:d}.pth'.format(epoch)
# ))
if (it + 1) % args.sample_interval == 0:
attgan.eval()
with torch.no_grad():
samples = [fixed_img_a]
for i, att_b in enumerate(sample_att_b_list):
att_b_ = (att_b * 2 - 1) * args.thres_int
if i > 0:
att_b_[..., i -
1] = att_b_[..., i -
1] * args.test_int / args.thres_int
samples.append(attgan.G(fixed_img_a, att_b_))
samples = torch.cat(samples, dim=3)
writer.add_image(
'sample',
vutils.make_grid(samples,
nrow=1,
normalize=True,
range=(-1., 1.)), it + 1)
vutils.save_image(samples,
os.path.join(
'output', args.experiment_name,
'sample_training',
'Epoch_({:d})_({:d}of{:d}).jpg'.format(
epoch, it % it_per_epoch + 1,
it_per_epoch)),
nrow=1,
img_unit = ((img_unit * 0.5) + 0.5) * 255
img_unit = np.uint8(img_unit)
img_unit = img_unit[::-1, :, :].transpose(1, 2, 0)
for i in range(args.n_attrs):
tmp = att_org.clone()
tmp[:, i] = 1 - tmp[:, i]
tmp = check_attribute_conflict(tmp, args.attrs[i], args.attrs)
att_list.append(tmp)
if args_.use_model == 'generator':
with torch.no_grad():
samples = [img_real]
for i, att_tar in enumerate(att_list):
if i > 0:
att_diff = att_tar - att_org
samples.append(attgan.G(img_real, att_diff))
samples = torch.cat(samples, dim=3)
out_file = '{:06d}.jpg'.format(idx)
vutils.save_image(samples,
join(output_path, out_file),
nrow=1,
normalize=True,
range=(-1., 1.))
print('{:s} done!'.format(out_file))
elif args_.use_model == 'discriminator':
with torch.no_grad():
result = img_unit
abn_att_real, cabn_att_real = attgan.D(img_real, 'abn')
for i in range(len(att_list) - 1):
abn_att = F.interpolate(abn_att_real[i],
size=(mw, mh),
'Face Attribute Manipulation') / 10
att_b[0][1] = cv2.getTrackbarPos('Bangs',
'Face Attribute Manipulation') / 10
att_b[0][2] = cv2.getTrackbarPos('Black_Hair',
'Face Attribute Manipulation') / 10
att_b[0][3] = cv2.getTrackbarPos('Blond_Hair',
'Face Attribute Manipulation') / 10
att_b[0][4] = cv2.getTrackbarPos('Brown_Hair',
'Face Attribute Manipulation') / 10
att_b[0][5] = cv2.getTrackbarPos('Bushy_Eyebrows',
'Face Attribute Manipulation') / 10
att_b[0][6] = cv2.getTrackbarPos('Eyeglasses',
'Face Attribute Manipulation') / 10
att_b[0][7] = cv2.getTrackbarPos('Male',
'Face Attribute Manipulation') / 10
att_b[0][8] = cv2.getTrackbarPos('Mouth_Slightly_Open',
'Face Attribute Manipulation') / 10
att_b[0][9] = cv2.getTrackbarPos('Mustache',
'Face Attribute Manipulation') / 10
att_b[0][10] = cv2.getTrackbarPos('No_Beard',
'Face Attribute Manipulation') / 10
att_b[0][11] = cv2.getTrackbarPos('Pale_Skin',
'Face Attribute Manipulation') / 10
att_b[0][12] = cv2.getTrackbarPos('Young',
'Face Attribute Manipulation') / 10
with torch.no_grad():
att_b_ = (att_b * 2 - 1) * args.thres_int
img_a = attgan.G(img_ori, att_b_)
_, att_test = attgan.D(img_a)
if (it + 1) % args.save_interval == 0:
# To save storage space, I only checkpoint the weights of G.
# If you'd like to keep weights of G, D, optim_G, optim_D,
# please use save() instead of saveG().
attgan.save(
os.path.join('output', args.experiment_name, 'checkpoint',
'weights.{:d}.pth'.format(epoch)))
if (it + 1) % args.sample_interval == 0:
attgan.eval()
with torch.no_grad():
samples = [fixed_img_real]
for i, att_trg in enumerate(sample_attr_list):
attr_diff = att_trg - sample_attr_list[0]
samples.append(attgan.G(fixed_img_real, attr_diff))
samples = torch.cat(samples, dim=3)
writer.add_image(
'sample',
vutils.make_grid(samples,
nrow=1,
normalize=True,
range=(-1., 1.)), it + 1)
vutils.save_image(samples,
os.path.join(
'output', args.experiment_name,
'sample_training',
'Epoch_({:d})_({:d}of{:d}).jpg'.format(
epoch, it % it_per_epoch + 1,
it_per_epoch)),
nrow=1,
attgan.eval()
for idx, (img_a, att_a) in enumerate(test_dataloader):
if args.num_test is not None and idx == args.num_test:
break
img_a = img_a.cuda() if args.gpu else img_a
att_a = att_a.cuda() if args.gpu else att_a
att_a = att_a.type(torch.float)
att_b_list = [att_a]
for i in range(args.n_attrs):
tmp = att_a.clone()
tmp[:, i] = 1 - tmp[:, i]
tmp = check_attribute_conflict(tmp, args.attrs[i], args.attrs)
att_b_list.append(tmp)
with torch.no_grad():
samples = [img_a]
for i, att_target in enumerate(att_b_list):
if i > 0:
att_diff = att_target - att_b_list[0]
samples.append(attgan.G(img_a, att_diff))
samples = torch.cat(samples, dim=3)
out_file = '{:06d}.jpg'.format(idx + 182638)
vutils.save_image(
samples, join(output_path, out_file),
nrow=1, normalize=True, range=(-1., 1.)
)
print('{:s} done!'.format(out_file))
