Cannot prep kitti dataset or train

[HukamiEstia]

Hi,

Ubuntu 18.04 python3.8 miniconda cuda 10.2 pytorch 1.5.1

I followed the instructions to install on the master branch (and spconv as well) but when running

python create_data.py kitti_data_prep /path/to/KITTI_DATASET_ROOT

I get a seg fault.

So I tried to find out where it happens, and I found that it is caused by import of function from second.core.non_max_suppression.nms_gpu

In the end I found that removing b[0] from the function trangle_area (misssing "i"?) allows the import and the script runs.

So I changed https://github.com/traveller59/second.pytorch/blob/3aba19c9688274f75ebb5e576f65cfe54773c021/second/core/non_max_suppression/nms_gpu.py#L169

to (a[0] - c[0]) * (b[1] - c[1]) - (a[1] - c[1])) / 2.0 and it runs successfully. However I have no idea what effect it has on the output...

Since I'm not sure it is even used I tried to train with:

python ./pytorch/train.py train --config_path=./configs/car.fhd.config --model_dir=/path/to/model_dir

It looks like it start but at some point I get another segmentation fault... and warnings from numba.

[patilunmesh]

@HukamiEstia I am facing the same issue of multiple segfaults. Did you get any solution to this?