search

HuguesTHOMAS / KPConv-PyTorch

Kernel Point Convolution implemented in PyTorch
MIT License
790 stars 158 forks source link

Problems with batch_num in pytorch version #56

Closed Zhen-ao closed 3 years ago

Zhen-ao commented 4 years ago

嗨，@HuguesTHOMAS, I am trying to use your library in the SemanticKitti dataset. I tried to modify the batch_num parameter from8 to 160 to make full use of computing resources. Unfortunately, the following error is reported at line 801 ofSemanticKitti.py:

self.dataset.epoch_inds += gen_indices
RuntimeError: The size of tensor a (35200) must match the size of tensor b (4071) at non-singleton dimension 0

I don’t know why there is such a problem when increasing batch_num, I can’t find a reasonable solution

HuguesTHOMAS commented 4 years ago

I need a little bit more information to know where the problem is. What line creates the error for example?

In the meantime you can look at this issue #42