LaserMix for Semi-Supervised LiDAR Semantic Segmentation

CVPR 2023  ·  Lingdong Kong, Jiawei Ren, Liang Pan, Ziwei Liu ·

Densely annotating LiDAR point clouds is costly, which restrains the scalability of fully-supervised learning methods. In this work, we study the underexplored semi-supervised learning (SSL) in LiDAR segmentation. Our core idea is to leverage the strong spatial cues of LiDAR point clouds to better exploit unlabeled data. We propose LaserMix to mix laser beams from different LiDAR scans, and then encourage the model to make consistent and confident predictions before and after mixing. Our framework has three appealing properties: 1) Generic: LaserMix is agnostic to LiDAR representations (e.g., range view and voxel), and hence our SSL framework can be universally applied. 2) Statistically grounded: We provide a detailed analysis to theoretically explain the applicability of the proposed framework. 3) Effective: Comprehensive experimental analysis on popular LiDAR segmentation datasets (nuScenes, SemanticKITTI, and ScribbleKITTI) demonstrates our effectiveness and superiority. Notably, we achieve competitive results over fully-supervised counterparts with 2x to 5x fewer labels and improve the supervised-only baseline significantly by 10.8% on average. We hope this concise yet high-performing framework could facilitate future research in semi-supervised LiDAR segmentation. Code is publicly available.

PDF Abstract CVPR 2023 PDF CVPR 2023 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Semi-Supervised Semantic Segmentation Cityscapes 12.5% labeled LaserMix (DeepLab v3+, ImageNet pre-trained ResNet50, single scale inference) Validation mIoU 77.1% # 7
Semi-Supervised Semantic Segmentation Cityscapes 25% labeled LaserMix (DeepLab v3+, ImageNet pre- trained ResNet50, single scale inference) Validation mIoU 78.3% # 9
Semi-Supervised Semantic Segmentation Cityscapes 50% labeled LaserMix (DeepLab v3+, ImageNet pre- trained ResNet50, single scale inference) Validation mIoU 79.1% # 8
Semi-Supervised Semantic Segmentation nuScenes LaserMix (Range View) mIoU (1% Labels) 49.5 # 4
mIoU (10% Labels) 68.2 # 2
mIoU (20% Labels) 70.6 # 2
mIoU (50% Labels) 73.0 # 2
Semi-Supervised Semantic Segmentation nuScenes LaserMix (Voxel) mIoU (1% Labels) 55.3 # 1
mIoU (10% Labels) 69.9 # 1
mIoU (20% Labels) 71.8 # 1
mIoU (50% Labels) 73.2 # 1
Semi-Supervised Semantic Segmentation ScribbleKITTI LaserMix (Range View) mIoU (1% Labels) 38.3 # 4
mIoU (10% Labels) 54.4 # 1
mIoU (20% Labels) 55.6 # 1
mIoU (50% Labels) 58.7 # 1
Semi-Supervised Semantic Segmentation ScribbleKITTI LaserMix (Voxel) mIoU (1% Labels) 44.2 # 1
mIoU (10% Labels) 53.7 # 2
mIoU (20% Labels) 55.1 # 2
mIoU (50% Labels) 56.8 # 2
Semi-Supervised Semantic Segmentation SemanticKITTI LaserMix (Range View) mIoU (1% Labels) 43.4 # 4
mIoU (10% Labels) 58.8 # 3
mIoU (20% Labels) 59.4 # 3
mIoU (50% Labels) 61.4 # 3
Semi-Supervised Semantic Segmentation SemanticKITTI LaserMix (Voxel) mIoU (1% Labels) 50.6 # 2
mIoU (10% Labels) 60.0 # 2
mIoU (20% Labels) 61.9 # 2
mIoU (50% Labels) 62.3 # 2

Methods


No methods listed for this paper. Add relevant methods here