tuple indices must be integers or slices, not str

Question

TypeError: tuple indices must be integers or slices, not str

    if args.cpool != 'no':
        # Forward into the network
        out = feat, out0, feat_att, out_att  = net(input1, input2)#修改，增加,dt_loss
        # Part attention loss
        loss_p = criterion1(out_att, labels)
        dt_loss = out['dt_loss']

Traceback (most recent call last):
File "E:/韩志忠学习/新运行/修改海军code/train.py", line 470, in
train(epoch)
File "E:/韩志忠学习/新运行/修改海军code/train.py", line 315, in train
dt_loss = out['dt_loss']
TypeError: tuple indices must be integers or slices, not str

loss0 = torch.mean(comask_pos * self.criterion0(feat, feat_recon_pos, feat_recon_neg))
return {'feat': feat_recon_pos, 'loss0': loss0}这是loss的返回值

怎么解决

Answer 1

报错的意思是 out 是一个元组，元组取值不能通过字符串形式：
dt_loss = out['dt_loss']
而要通过 索引形式
例如 out[0]
根据你的代码取就好了