def configure():
args = parse_arguments()
configure_dataset(args)
configure_paths(args)
args.scales = [args.initial_scale]
for i in range(1, args.n_scales):
args.scales.append(args.scales[-1] * args.scale_step)
args.arch = args.model
with (args.result_path / 'opts.json').open('w') as opt_file:
json.dump(vars(args), opt_file, default=json_serialize)
args.tee = TeedStream(args.result_path / "output.log")
args.time_suffix = datetime.now().strftime("%d%m%H%M")
tb_path = args.result_path / "tb"
args.writer = SummaryWriter(tb_path.as_posix())
args.device = torch.device("cuda" if args.cuda else "cpu")
create_code_snapshot(
Path(__file__).parent, args.result_path / "snapshot.tgz")
torch.manual_seed(args.manual_seed)
args.logger = setup_logging(args)
return args
def main():
args = parse_arguments()
net, _ = create_model(args, args.model)
net = net.module
net.cpu()
if net is None:
return
net.eval()
h, w = args.sample_size, args.sample_size
var = torch.randn(1, args.sample_duration, 3, h, w).to('cpu')
net.apply(lambda m: m.register_forward_hook(compute_layer_statistics_hook))
out = net(var)
restore_module_names(net)
print_statisctics()
