yanx27 / JS3C-Net

Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completion (AAAI 2021)
MIT License
211 stars 34 forks source link

Semantic Scene Completion GT #4

Closed superxiaoying closed 3 years ago

superxiaoying commented 3 years ago

Hi,

Thanks for your great work.

I want to know that the SSC GT label "data_odometry_voxels_all.zip" is from the semantic-kitti official or you prepared it ? http://semantic-kitti.org/assets/data_odometry_voxels_all.zip

Best, Ying

yanx27 commented 3 years ago

@superxiaoying Hi, thanks for your attention. We use official SSC GT from semantickitti while use mannually generated GT for semanticposs dataset.

superxiaoying commented 3 years ago

@yanx27 Thx. I think there maybe a little unfair for SSC task, since the number of training samples for standard SSC is about 1/5 of the all. The reported performance of JS3CNet paper uses all data, right? Have you tried less training samples and how much influence will it cause?

Best, Ying

yanx27 commented 3 years ago

Yes, we used 1/5 data for training of SSC, and it can achieve similar results. I think this because of the redundancy of LiDAR sequences.

superxiaoying commented 3 years ago

Thanks for your response.