参考文章:https://blog.csdn.net/weixin_42028364/article/details/81675021
collate_fn是torch.utils.data.DataLoaderd的参数之一,用来定义如何取样,可以不赋值,也可以自己定义函数来实现自己想要的功能。
比如:
def collate_fn(data):
images,captions, sentence_num, max_word_num, image_id = zip(*data)
images = torch.stack(images, 0)
max_sentence_num = max(sentence_num)
max_word_num = max(max_word_num)
prob = np.zeros((len(captions), max_sentence_num + 1))
for i, caption in enumerate(captions):
for j, sentence in enumerate(caption):
prob[i][j] = len(sentence) > 0
return images, targets, prob, image_id
data_loader = torch.utils.data.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=shuffle,
collate_fn=collate_fn,
num_workers=num_workers)