def initrun(dataloader, netD, netG, args):
# model train or inference
if args.mode in ['trainsingle', 'trainrandom']:
mes = train_single_DRGAN(dataloader, netD, netG, args)
if args.mode=='trainmulti':
mes = train_multiple_DRGAN(dataloader, netD, netG, args)
if args.mode in ['gensingle', 'genmulti']:
mes = generate_image(dataloader, netG, args)
if args.mode in ['idensingle', 'idenmulti']:
mes = representation_learning(dataloader, netG, args)
return mes
else:
D = multi_model.Discriminator(Nd, Np, channel_num)
G = multi_model.Generator(Np, Nz, channel_num,
args.images_perID)
else:
print('\nLoading model from [%s]...' % args.snapshot)
try:
D = torch.load('{}_D.pt'.format(args.snapshot))
G = torch.load('{}_G.pt'.format(args.snapshot))
except:
print("Sorry, This snapshot doesn't exist.")
exit()
if not (args.generate):
if not (args.multi_DRGAN):
train_single_DRGAN(images, id_labels, pose_labels, Nd, Np, Nz, D,
G, args)
else:
if args.batch_size % args.images_perID == 0:
train_multiple_DRGAN(images, id_labels, pose_labels, Nd, Np,
Nz, D, G, args)
else:
print(
"Please give valid combination of batch_size, images_perID"
)
exit()
else:
# pose_code = [] # specify arbitrary pose code for every image
pose_code = np.random.uniform(-1, 1, (images.shape[0], Np))
features = Generate_Image(images, pose_code, Nz, G, args)