def load_net(self, net_name, set_net_version=None):
net_path = os.path.join(self.trained_nets_path, net_name, 'model_dir', 'dynamic_net.pth')
temp_opt = config.get_configurations()
opt_path = os.path.join(self.trained_nets_path, net_name, 'config.txt')
if os.path.exists(opt_path):
opt = utils.read_config_and_arrange_opt(opt_path, temp_opt)
else:
opt = temp_opt
self.main_widget.dynamic_model = InferenceModel(opt, set_net_version=set_net_version)
self.main_widget.dynamic_model.load_network(net_path)
parser.add_argument('--num_of_images', default=num_of_images, type=int)
inference_opt = parser.parse_args()
network_name = inference_opt.network_name
num_of_images = inference_opt.num_of_images
networks_path = os.path.join('trained_nets', network_name)
model_path = os.path.join(networks_path, 'model_dir', 'dynamic_net.pth')
config_path = os.path.join(networks_path, 'config.txt')
save_path = os.path.join('results', 'inference_results')
if not os.path.exists(save_path):
utils.make_dirs(save_path)
opt = config.get_configurations(parser=parser)
if os.path.exists(config_path):
utils.read_config_and_arrange_opt(config_path, opt)
dynamic_model = InferenceModel(opt)
dynamic_model.load_network(model_path)
dynamic_model.net.train()
to_tensor = transforms.ToTensor()
to_pil_image = transforms.ToPILImage()
first_image = True
input_tensor = torch.randn((128, dynamic_model.opt.z_size)).view(-1, dynamic_model.opt.z_size, 1, 1).to(dynamic_model.device)
for alpha in tqdm(alphas):
output_tensor = dynamic_model.forward_and_recover(input_tensor.requires_grad_(False), alpha=alpha)
image_tensor = torchvision.utils.make_grid(output_tensor[:num_of_images, :, :, :].clamp(min=0.0, max=1), nrow=1)
if first_image: