Unable to Run

[swarmt]

Results look excellent however unfortunately I am unable to run.

OS: Ubuntu 18.04
GPU: 1050ti

python3 ./pytorch/train.py evaluate --config_path=./configs/car.config --model_dir=pretrained_models/car/

Results in Segmentation fault (core dumped)

pip3 freeze

[traveller59]

1. you need to change batch size to 1 in config file because 1050ti only have 4gb memory.
2. use gdb or pdb to find problem location, I can't find problem in my environment, please provide more information.

[traveller59]

Do you modify files in SparseConvNet? I think your problem may due to this. pretrained model doesn't support SparseConvNet master, you need to follow README.md to modify some line after you install SparseConvNet.

[swarmt]

That could be the issue.

Installing in a conda env I get the error.

g++ /tmp/tmp5yesv3u8.cc -o tmp5yesv3u8 -I/usr/include -L/usr/lib64 -lcudart /tmp/tmp5yesv3u8.cc:4:23: error: ‘::main’ must return ‘int’ void main(){ ^ g++ /tmp/tmpvyirytjp.cc -o tmpvyirytjp -I/usr/include -L/usr/lib64 -lcudart /tmp/tmpvyirytjp.cc:4:23: error: ‘::main’ must return ‘int’ void main(){ ^ Traceback (most recent call last): File "/home/code/second.pytorch/second/core/non_max_suppression/nms_cpu.py", line 10, in <module> from second.core.non_max_suppression.nms import ( ModuleNotFoundError: No module named 'second.core.non_max_suppression.nms'

Would you like me to create an installation script to replicate the install steps exactly? Might need to specify the pip versions as well.

[traveller59]

@swarmt g++ error should be a problem during finding cuda arch, see this.. I have change the code to int main()... and fixed a serious bug, remember to update the code.
No, the install step only contains three step: 1) install dependencies. 2) install SparseConvNet. 3) prepare dataset.

[swarmt]

Excellent. I think we can call it fixed.