IndexError: too many indices for tensor of dimension 4
HCookY95 opened this issue · comments
@Open-Debin
求问答主，我用您的model resnet18_FAN，有这样的报错是怎么回事？
f = x[:, :, :, :, i]
IndexError: too many indices for tensor of dimension 4
Hi,
The tensor is shaped in [batch, channel, height, width, N_segment ]
The code is designed for processing video data, where the input of the FAN model is N_segment x frames, each frame is a 4-dimensional tensor, therefore N_segment x frames is a 5-dimensional tensor. If you input a 4-dimensional tensor, the error must happen.
Hi,
The tensor is shaped in [batch, channel, height, width, N_segment ]
The code is designed for processing video data, where the input of the FAN model is N_segment x frames, each frame is a 4-dimensional tensor, therefore N_segment x frames is a 5-dimensional tensor. If you input a 4-dimensional tensor, the error must happen.
请问作者有邮箱或者微信吗，能否私下询问一下？
Hi,
The tensor is shaped in [batch, channel, height, width, N_segment ]
The code is designed for processing video data, where the input of the FAN model is N_segment x frames, each frame is a 4-dimensional tensor, therefore N_segment x frames is a 5-dimensional tensor. If you input a 4-dimensional tensor, the error must happen.
@Open-Debin
我删掉了num_pair=3的参数，在train中只执行一次，就没有这个报错了。
但是执行自己的main file以后出现了NoneType has no attribute data。这是我自己定义的一个梯度裁剪的function。
def clip_gradient(optimizer, grad_clip): for group in optimizer.param_groups: for param in group['params']: param.grad.data.clamp_(-grad_clip, grad_clip) print (param.grad)
然后我print了一下每层的weight和bias，发现self.beta和self.pred_fc2这两层没有参数，没法进行反向传播，所以直接在ResNet_AT里comment了这两层，请问这样做是对的吗，还是因为这两层在network里detached了，我应该用什么方式连进去？
for name, param in model.named_parameters(): print(name, param.grad)
另外，没有再报错了，不过CK+的准确率卡在了93-94之间，是因为上面提到的问题，还是我应该尝试一些其他的lr scheduler？
@HCookY95
Merry Christmas! I recently update the Emotion-FAN, new features include data process, environment install, CK+ code, and Baseline code. Also, you can find the old version directory of Emotion-FAN in the README.md. I hope my new updates can help you greatly. Please see the Emotion-FAN for more details.