Open AppleAndBanana opened 4 years ago
Trying to use nn.functional.sigmoid instead of nn.Sigmoid can solve this problem.
However, the _tanh()
still have some bugs, cause it will generate two output blobs like this.
The reson of this situation is pytorch_to_caffe.py
has two duplicate function of F.tanh = Rp(F.tanh, _tanh)
. Comment one of them can solve the problem.
Mark,调试了半天,终于找到问题解决了!感谢感谢!!
I found that the function '_sigmoid() ' and '_tanh()' in pytorch_to_caffe.py has no return. This will result in failure when my net has sigmoid or tanh module. Maybe is this a bug?