Research map / Detection and segmentation / 3D detection and driving
Robo3D: Towards Robust and Reliable 3D Perception against Corruptions
Lingdong Kong, Youquan Liu, Xin Li, Runnan Chen, Wenwei Zhang, Jiawei Ren, Liang Pan, Kai Chen, Ziwei Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| state-of-the-art segmentors with higher clean accuracy versus the MinkUNet baseline | recent 3D segmentors (2DPASS, GFNet, RPVNet, CPGNet) · SemanticKITTI-C | mCE: 2DPASS 106.1, GFNet 108.7 vs MinkUNet 100.0 | Hurt |
| use projection-based range-view or BEV representation instead of voxel/point-voxel | LiDAR segmentors · SemanticKITTI-C | mCE: SqueezeSeg 164.9, CENet 103.4 vs SPVCNN-34 99.2 | Hurt |
| use out-of-context augmentation (mixing scenes) instead of in-context augmentation (ablation) | baseline segmentor · Robo3D segmentation | mCE: 100.0 -> 86.0 | Improved |
| add flexible (randomized) voxelization (ablation) | baseline with ICA · Robustness training (with ICA) | mCE: 100.0 -> 97.4 | Improved |
| add completion loss (part2full) on top of flexible voxelization (ablation) | density-insensitive training with ICA · Robustness training (with ICA) | mCE: 97.4 -> 96.4 | Improved |
| add confirmation loss (full2part) in addition to completion loss (ablation) | density-insensitive training with ICA · Robustness training (with ICA) | mCE: 96.4 -> 96.1 | Improved |
| add flexible voxelization to OCA-trained model (ablation) | baseline with OCA · Robustness training (with OCA) | mCE: 86.0 -> 84.5 | Improved |
| add both consistency losses to OCA-trained model (ablation) | baseline with OCA · Robustness training (with OCA) | mRR: 84.7 -> 89.7 | Improved |
| apply flexible voxelization and density-insensitive training (ablation) | baseline with OCA · Robustness training (with OCA) | clean accuracy: 69.2 -> 68.1 | Hurt |
| apply flexible voxelization and density-insensitive training (ablation) | baseline with ICA · Robustness training (with ICA) | clean accuracy: 62.8 -> 62.7 | No clear effect |
| raise mask ratio beyond range of 0.3 to 0.6 (ablation) | density-insensitive training · Robo3D | mCE | Hurt |
| use larger versus smaller fixed voxel size (ablation) | baseline segmentor · Robo3D | mCE | Mixed |
Closest papers by meaning
- RoboDepth: Robust Out-of-Distribution Depth Estimation under Corruptions (NeurIPS 2023 · similarity 0.89)
- RCP-Bench: Benchmarking Robustness for Collaborative Perception Under Diverse Corruptions (CVPR 2025 · similarity 0.87)
- Scalable Benchmarking and Robust Learning for Noise-Free Ego-Motion and 3D Reconstruction from Noisy Video (ICLR 2025 · similarity 0.86)
- OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection (NeurIPS 2025 · similarity 0.85)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (CVPR 2024 · similarity 0.85)
- Cocoon: Robust Multi-Modal Perception with Uncertainty-Aware Sensor Fusion (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
