你好，我这地方不太理解，您方便解答一下吗？

[liubaoning111]

此处的dataset=labeled_train_set引用的是有标记的数据，但是dataset=train_set却还是总的数据集，为什么不是unlabeled_train_set呢？

[liubaoning111]

train_set = ISICDataset(image_path=args.data_path, stage='train', image_size=args.image_size, is_augmentation=True) #1045
labeled_train_set, unlabeled_train_set = random_split(train_set, [int(len(train_set) * args.labeled_percentage),   ## 1045,  52, 
                                                   len(train_set) - int(len(train_set) * args.labeled_percentage)]) # 1045,  993个无标记的样本 
print('before:', len(labeled_train_set), len(train_set))   #52个有标记的样本  
# repeat the labeled set to have a equal length with the unlabeled set (dataset)
labeled_ratio = len(train_set) // len(labeled_train_set)  ##20倍
labeled_train_set = ConcatDataset([labeled_train_set for i in range(labeled_ratio)])  #翻倍20次  变成1040
labeled_train_set = ConcatDataset([labeled_train_set,
                                   Subset(labeled_train_set, range(len(train_set) - len(labeled_train_set)))])
assert len(labeled_train_set) == len(train_set)
print('after:', len(labeled_train_set), len(train_set))
train_labeled_dataloder = DataLoader(dataset=labeled_train_set, num_workers=args.num_workers, batch_size=args.batch_size, shuffle=True, pin_memory=True)
train_unlabeled_dataloder = DataLoader(dataset=train_set, num_workers=args.num_workers, batch_size=args.batch_size, shuffle=True, pin_memory=True)是这些代码中的

[Senyh]

将全集作为无标注数据集，使一致性正则能够利用全部数据