Closed cjyiiiing closed 3 years ago
It is strange and I think it is an issue of spconv; You can refer to this issue
It is strange and I think it is an issue of spconv; You can refer to this issue
I read this issue and find no solution to this problem. I print out the input and output of conv3 in UpBlock. There is no nan. Does the nan problem affect the training?
From empirical study, it does not.
Thanks!
I download the pretrained model for SemanticKITTI and print all parameters. Then I find the weight of conv3 in UpBlock always exist NAN like this:
I'm very confused about it. Could you explain why?