Cross-modal and Cross-domain Knowledge Transfer for Label-free 3D Segmentation
Not on the map. Why?
Abstract (arXiv)
Current state-of-the-art point cloud-based perception methods usually rely on large-scale labeled data, which requires expensive manual annotations. A natural option is to explore the unsupervised methodology for 3D perception tasks. However, such methods often face substantial performance-drop difficulties. Fortunately, we found that there exist amounts of image-based datasets and an alternative can be proposed, i.e., transferring the knowledge in the 2D images to 3D point clouds. Specifically, we propose a novel approach for the challenging cross-modal and cross-domain adaptation task by fully exploring the relationship between images and point clouds and designing effective feature alignment strategies. Without any 3D labels, our method achieves state-of-the-art performance for 3D point cloud semantic segmentation on SemanticKITTI by using the knowledge of KITTI360 and GTA5, compared to existing unsupervised and weakly-supervised baselines.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 30 of its 37 references that are in the map. First posted on arXiv 19 Sep 2023.
Builds on
30 of 37 references.
References · 30
- 2014Adam: A Method for Stochastic Optimizationcited by 37,240
- 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
- 2022Generative Adversarial Networkscited by 5,627
- 2016DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFscited by 3,241
- 2018Dynamic Graph CNN for Learning on Point Cloudscited by 2,223
- 2016Playing for Data: Ground Truth from Computer Gamescited by 818
- 2018ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentationcited by 623
- 2021KITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3Dcited by 575
- 2019Deep Learning for 3D Point Clouds: A Surveycited by 558
- 2020Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentationcited by 389
- 2021Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentationcited by 268
- 2017SqueezeSeg: Convolutional Neural Nets with Recurrent CRF for Real-Time Road-Object Segmentation from 3D LiDAR Point Cloudcited by 232
18 more
- 2020SalsaNext: Fast, Uncertainty-aware Semantic Segmentation of LiDAR Point Clouds for Autonomous Drivingcited by 226
- 2020DACS: Domain Adaptation via Cross-domain Mixed Samplingcited by 226
- 20222DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point Cloudscited by 158
- 2019xMUDA: Cross-Modal Unsupervised Domain Adaptation for 3D Semantic Segmentationcited by 129
- 2022Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Datacited by 126
- 2022Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentationcited by 119
- 2021One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentationcited by 89
- 2021DANNet: A One-Stage Domain Adaptation Network for Unsupervised Nighttime Semantic Segmentationcited by 80
- 2022SIGMA: Semantic-complete Graph Matching for Domain Adaptive Object Detectioncited by 74
- 2022Scribble-Supervised LiDAR Semantic Segmentationcited by 69
- 2021Category Contrast for Unsupervised Domain Adaptation in Visual Taskscited by 65
- 2020Domain Adaptive Semantic Segmentation Using Weak Labelscited by 56
- 2022LESS: Label-Efficient Semantic Segmentation for LiDAR Point Cloudscited by 41
- 2020TORNADO-Net: mulTiview tOtal vaRiatioN semAntic segmentation with Diamond inceptiOn modulecited by 36
- 20213D Spatial Recognition without Spatially Labeled 3Dcited by 31
- 2022CL3D: Unsupervised Domain Adaptation for Cross-LiDAR 3D Detectioncited by 27
- 2022SimT: Handling Open-set Noise for Domain Adaptive Semantic Segmentationcited by 17
- 2022Unsupervised Domain Adaptation for Point Cloud Semantic Segmentation via Graph Matchingcited by 7
Built on by
Citing papers per year: 2024: 1.
1 paper.