Paper accepted at ICRA2023 – COLA: COarse LAbel pre-training for 3D semantic segmentation of sparse LiDAR datasets

https://www.caor.minesparis.psl.eu/wp-content/uploads/2023/02/1674205575126-500x400.jpg

Congratulations to Jules Sanchez for his paper accepted at #ICRA2023

How to improve 3D semantic segmentation performance using more 3D data? Pre-training with "coarse" labels. This requires fewer raw data than typical contrastive pre-training approaches and can really boost performance when the target dataset has few annotations!

Paper on arXiv: https://arxiv.org/pdf/2202.06884.pdf

#lidar #semanticsegmentation #pretrain #autonomousdriving #ICRA2023