xinge008 / Cylinder3D

Rank 1st in the leaderboard of SemanticKITTI semantic segmentation (both single-scan and multi-scan) (Nov. 2020) (CVPR2021 Oral)
Apache License 2.0
859 stars 180 forks source link

Why calculate IoU from voxel label? #63

Closed jialeli1 closed 3 years ago

jialeli1 commented 3 years ago

Hi.

I a new to point cloud segmentation. There are problems that have been bothering me a lot.

1). Why do you (here) and PolarNet (here) both calculate IoU from voxel-wise label instead of point-wise label? Even PolarNet directly saves the voxel label to the file for submission (here).

2.) As I learned from semantic-kitti-api, predictions about each point of the scan are required when submitted to the benchmark. Is my understanding correct? Or is it not necessary to predict the label of every point?

3.) If the prediction at each point is required, then another problem arises. You clamped the points in the fixed_volume_space here, thus your voxel-wise predictions also should only appear in the fixed_volume_space. However, I tried to load some point cloud samples, the range of point distribution is much larger than this setting, and even reached (-80.m, -80.0m, -10.0m, +80.0m, +80.0m, +3.0m) for x, y, z-axis. How to determine the predictions of these out-of-range points if necessary?

Thank you for answering me. It really bothers me very much.