print('==> Training', args.student if args.name is None else args.name)
print('==> Building model..')
# get models as teachers and students
teachers, student = get_model(args, config, device="cuda")
print("==> Teacher(s): ", " ".join([teacher.__name__ for teacher in teachers]))
print("==> Student: ", args.student)
dims = [student.out_dims[i] for i in eval(args.out_layer)]
print("dims:", dims)
update_parameters = [{'params': student.parameters()}]
if args.adv:
discriminators = discriminator.Discriminators(dims, grl=args.grl)
for d in discriminators.discriminators:
d = d.to(device)
if device == "cuda":
d = torch.nn.DataParallel(d)
update_parameters.append({'params': d.parameters(), "lr": args.d_lr})
print(args)
if args.resume:
# Load checkpoint.
print('==> Resuming from checkpoint..')
assert os.path.isdir('checkpoint'), 'Error: no checkpoint directory found!'
checkpoint = torch.load(
'/workspace/mnt/storage/yangdecheng/yangdecheng/models/checkpoint/%s-generator/ckpt.t7'
% "_".join(args.teachers))
print('==> Training', args.student if args.name is None else args.name)
print('==> Building model..')
# get models as teachers and students
teachers, student = get_model(args, config, device="cuda")
print("==> Teacher(s): ", " ".join([teacher.__name__ for teacher in teachers]))
print("==> Student: ", args.student)
dims = [student.out_dims[i] for i in eval(args.out_layer)]
print("dims:", dims)
update_parameters = [{'params': student.parameters()}]
if args.adv:
discriminators = discriminator.Discriminators(dims)
for d in discriminators.discriminators:
d = d.to(device)
if device == "cuda":
d = torch.nn.DataParallel(d)
update_parameters.append({'params': d.parameters(), "lr": args.d_lr})
print(args)
if args.resume:
# Load checkpoint.
print('==> Resuming from checkpoint..')
assert os.path.isdir('checkpoint'), 'Error: no checkpoint directory found!'
checkpoint = torch.load('./checkpoint/%s-generator/ckpt.t7' % "_".join(args.teachers))
student.load_state_dict(checkpoint['net'])
start_epoch = checkpoint['epoch']