dgcnn cuda extension in training and evaluation

[bnojavan]

I am trying this framework on Ubuntu 18.04 and following the instructions provided. When I was compiling the cuda extension I was getting "TH_INDEX_BASE" is undefined. I manually defined it in the c source code to be 1 following https://github.com/torch/torch7/blob/master/lib/TH/THGeneral.h.in and was able to "seemingly" build the cuda extension and it generated the build folder. However, I still see " Please compile source files before using dgcnn cuda extension." When I am trying to train/test, although there are no errors, it gets to an infinite loop scanning all images in one folder of the test and does not end running.

[callmeray]

I think it's related to the version of PyTorch. Try downgrade to pytorch1.0.1 and compile again.

[FishHe]

Same problem I have met.
One can easily fix the bug by replacing TH_Index_Base with 0.
The official solution: pytorch/pytorch#17591.

[Innocence4822]

Hello, I'm sorry to bother you. Can you tell me where I can find th_ Index_ Base？ @FishHe

[Innocence4822]

When i run the test.py ,An error occurred “NameError: name 'dgcnn_ext' is not defined” . But ,I have run "bash compile.sh“ and succeeded. I don't know how to solve the problem.Can you help me?@callmeray

[freyang]

Hello, I'm sorry to bother you. Can you tell me where I can find th_ Index_ Base？ @FishHe

gather_knn_kernel.cu

[Innocence4822]

Sorry, I still can't understand. Can you explain it in detail？😊 @HelenYang1999

[geviolet]

你好，我最近运行这个代码也出现和你一样的问题，请问您解决了吗？可否将解决之法告知