tyagi-iiitv / PointPillars

GNU General Public License v3.0
105 stars 47 forks source link

Loss becoming inf immediately for size / loc #47

Open kristiangulakerkm opened 2 years ago

kristiangulakerkm commented 2 years ago

I am training on a custom dataset, and notice while all other loss functions return normal values, size/loc are inf immediately. The point clouds I am training on are min-max normalized, but I have done no other preprocessing steps. Any guidance you can give me? I have tried playing with config values however I see little effect on the problem.