what about using more point in input cloud, 80000, 100000 and so on ?

facebookresearch / votenet

Deep Hough Voting for 3D Object Detection in Point Clouds

MIT License

1.67k stars 377 forks source link

what about using more point in input cloud, 80000, 100000 and so on ? #120

Open xinghuokang opened 3 years ago

xinghuokang commented 3 years ago

More points, better detection result?

ljx398966986 commented 3 years ago

I think there will be more wrong bboxs, you can compare 20000 and 40000

xinghuokang commented 3 years ago

why? If we train the network with more point, not 20000, will the result be better? Moreever, sample less point may loss many information, especially the input point is sparse.