def main(): root = r"/media/root/文档/wqr/my_spixel_fcn/data_preprocessing/up_facet" dataset = Data_api_rec(root=root) from data_api.get_dataset_mean_std import get_mean_std a, b, _ = dataset[0] mean, std = get_mean_std(dataset, ratio=0.1) # print(mean) # print(std) # exit() # print(a.shape) # print(b.shape) from torchvision import transforms tran = transforms.ToPILImage() imga = tran(a) imgb = tran(b) imgaa = cv2.cvtColor(np.array(imga), cv2.COLOR_RGB2BGR) cv2.imshow("a", np.uint8(imgaa)) cv2.imshow("b", np.uint8(imgb)) cv2.waitKey()
test_parser.add_argument("--continue_train", "-ct", default=False, type=bool, help="continue train using trained model") test_parser.add_argument( "--state_path", "-sp", default="checkpoints/network_state/acc94.74_model.pth", help="the p500ath of trained model") opt = test_parser.parse_args() print(opt.device) # 获取数据集的均值、方差 Norm, [mean, std] = get_mean_std(mean_std_path=os.path.join(base_path, "data_api")) test_data = Data_image_label(root=opt.data_root, updata_txt=True, train_test_ratio=0.0, phase="test", Norm=Norm) test_loader = DataLoader(test_data, batch_size=opt.batch_size, shuffle=False) # model model = Res18_UNet(n_classes=2, layer=4).to(opt.device) # device_ids = [0] # model = torch.nn.DataParallel(model, device_ids=device_ids) # model = model.cuda(device=device_ids[0]) model.eval()
train_parser.add_argument("--state_path", "-sp", default="checkpoints/network_state/network_epo1.pth", help="the path of trained model") opt = train_parser.parse_args() print(opt.device) train_data = Data_api_rec(root=opt.data_root, gen_label=False, updata_txt=False, train_test_ratio=0.8, phase="train", use_relative_path=False) Norm, [mean, std] = get_mean_std(mean_std_path=os.path.join(base_path, "data_api"), dataset=train_data) train_data = Data_api_rec(root=opt.data_root, gen_label=False, updata_txt=False, train_test_ratio=0.8, phase="train", use_relative_path=False, Norm=Norm) val_data = Data_api_rec(root=opt.data_root, gen_label=False, updata_txt=False, train_test_ratio=0.8, phase="val", Norm=Norm)