Research map / Point clouds / Point cloud classification
3D Test-time Adaptation via Graph Spectral Driven Point Shift
Xin Wei, Qin Yang, Yijie Fang, Mingrui Zhu, Nannan Wang
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 |
|---|---|---|---|
| add graph spectral driven point shift (GSDPS) input adaptation (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) | mean accuracy (%): +4.55% | Improved |
| add graph spectral guided model adaptation (GSGMA) (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) | mean accuracy (%): +3.48% | Improved |
| add outlier-aware graph (gamma>0) on background corruption (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) background | mean accuracy (%): 18.24% -> 69.54% | Improved |
| add outlier-aware graph on other 14 corruptions (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) other 14 corruptions | mean accuracy (%): 61.28 vs 61.44 | No clear effect |
| use eigenmap-guided pseudo-labels instead of deep-feature-only labels (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) | mean accuracy (%): 0.63% | Improved |
| vary loss weights beta1, beta2, beta3 (ablation) | GSDTTA · ScanObjectNN-C (DGCNN) shear | mean accuracy (%): standard deviations of 0.27, 0.25, and 0.95 | No clear effect |
| GSDTTA versus best 2D TTA method SHOT | GSDTTA · ModelNet40-C | mean accuracy (%): +1.71%, +1.39%, +2.05% | Improved |
| GSDTTA versus 3D TTA method CloudFixer | GSDTTA · ModelNet40-C | mean accuracy (%): +2.53% (DGCNN), +4.72% (CurveNet), +6.47% (PointNeXt) | Improved |
| GSDTTA versus CloudFixer | GSDTTA · ScanObjectNN-C | mean accuracy (%): +1.10%, +3.63%, +1.83% | Improved |
| GSDTTA versus CloudFixer on noise corruptions (uniform, gaussian, impulse, upsampling) | GSDTTA · ScanObjectNN-C | mean accuracy (%): average margin of 8.35% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CloudFixer — CloudFixer: Test-Time Adaptation for 3D Point Clouds via Diffusion-Guided Geometric Transformation (ECCV 2024) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CloudFixer: Test-Time Adaptation for 3D Point Clouds via Diffusion-Guided Geometric Transformation (ECCV 2024 · similarity 0.88)
- PC-Adapter: Topology-Aware Adapter for Efficient Domain Adaption on Point Clouds with Rectified Pseudo-label (ICCV 2023 · similarity 0.88)
- Exploring the Devil in Graph Spectral Domain for 3D Point Cloud Attacks (ECCV 2022 · similarity 0.87)
- SMART-PC: Skeletal Model Adaptation for Robust Test-Time Training in Point Clouds (ICML 2025 · similarity 0.86)
- Improving Robustness of 3D Point Cloud Recognition from a Fourier Perspective (NeurIPS 2024 · similarity 0.86)
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 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.
