train_mean = [0.485, 0.456, 0.406, 0.5] train_std = [0.229, 0.224, 0.225, 0.25] test_mean = [0.485, 0.456, 0.406, 0.5] test_std = [0.229, 0.224, 0.225, 0.25] # 配置transform # 注意:train和val涉及到label,需要用带_DL后缀的transform # test不涉及label,用原来的transform prob = 0.5 train_transform = T.Compose([ T_DL.ToTensor_DL(), # 转为tensor T_DL.RandomFlip_DL(p=prob), # 概率p水平或者垂直翻转 T_DL.RandomRotation_DL(p=prob), # 概率p发生随机旋转(只会90,180,270) # T_DL.RandomColorJitter_DL(p=prob, brightness=1, contrast=1, saturation=1, hue=0.5), # 概率p调整rgb T_DL.Normalized_DL(mean=train_mean[:input_channel], std=train_std[:input_channel]), # 归一化 ]) val_transform = T.Compose([ T_DL.ToTensor_DL(), # 转为tensor T_DL.Normalized_DL(mean=train_mean[:input_channel], std=train_std[:input_channel]), # 归一化 ]) test_transform = T.Compose([ # T.ToTensor(), T.Normalize(mean=test_mean[:input_channel], std=test_std[:input_channel]), ]) dataset_cfg = dict( # dir全都改成list
train_mean = [0.485, 0.456, 0.406, 0.5] train_std = [0.229, 0.224, 0.225, 0.25] test_mean = [0.485, 0.456, 0.406, 0.5] test_std = [0.229, 0.224, 0.225, 0.25] # 配置transform # 注意:train和val涉及到label,需要用带_DL后缀的transform # test不涉及label,用原来的transform prob = 0.25 train_transform = T.Compose([ T_DL.ToTensor_DL(), # 转为tensor T_DL.RandomFlip_DL(p=prob), # 概率p水平或者垂直翻转 # T_DL.RandomRotation_DL(p=prob), # 概率p发生随机旋转(只会90,180,270) T_DL.RandomColorJitter_DL(p=prob, brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5), # 概率p调整rgb T_DL.Normalized_DL(mean=train_mean[:input_channel if input_channel != 4 else -1], std=train_std[:input_channel if input_channel != 4 else -1]), # 归一化 ]) val_transform = T.Compose([ T_DL.ToTensor_DL(), # 转为tensor T_DL.Normalized_DL(mean=train_mean[:input_channel if input_channel != 4 else -1], std=train_std[:input_channel if input_channel != 4 else -1]), # 归一化 ]) test_transform = T.Compose([ T.ToTensor(), T.Normalize(mean=test_mean[:input_channel if input_channel != 4 else -1], std=test_std[:input_channel if input_channel != 4 else -1]), ]) dataset_cfg = dict( # train_dir=root_dir + '/tcdata/suichang_round1_train_210120',