Research map / Label-efficient and robust learning / Test-time adaptation
Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token purging
Moslem Yazdanpanah, Ali Bahri, Mehrdad Noori, Sahar Dastani, Gustavo Adolfo Vargas Hakim, David Osowiechi, Ismail Ben Ayed, Christian Desrosiers
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 BatchNorm reset (ablation) | PG-SP · ScanObjectNN-C mean | Top-1 accuracy (%): 39.86 -> 49.13 | Improved |
| add BatchNorm reset (ablation) | BFTT3D · ScanObjectNN-C mean | Top-1 accuracy (%): 33.00 -> 45.76 | Improved |
| increase purge size up to about 100 tokens (ablation) | PG · ScanObjectNN-C Background | Top-1 accuracy (%) | Improved |
| excessive purging (purge size above about 100) (ablation) | PG · ScanObjectNN-C Background | Top-1 accuracy (%) | Hurt |
| select purge size by entropy over six candidates {0,2,4,8,16,32} (ablation) | PG · ScanObjectNN-C Distortion | Top-1 accuracy (%): 59 -> 62.48 | Improved |
| purge tokens (PG-SP) instead of BFTT3D backprop-free adaptation | PG-SP · ScanObjectNN-C mean | Top-1 accuracy (%): 33.00 -> 49.13 | Improved |
| apply PG-SP to real-world ScanObjectNN OBJ-BG variant | PG-SP · ScanObjectNN OBJ-BG | Top-1 accuracy (%): 74.18 -> 75.56 | Improved |
| apply PG-SP to real-world ScanObjectNN PB-T50-RS variant | PG-SP · ScanObjectNN PB-T50-RS | Top-1 accuracy (%): 56.77 -> 61.07 | Improved |
| apply PG-SP to Point-Mamba backbone | PG-SP · ScanObjectNN-C mean (Mamba) | Top-1 accuracy (%): 29.11 -> 41.42 | Improved |
| apply PG-SP to Point-Mamba backbone | PG-SP · ShapeNet-C mean (Mamba) | Top-1 accuracy (%): 51.92 -> 51.66 | No clear effect |
| use PG-SP instead of BFTT3D for compute | PG-SP · GPU time per batch | ms: 173.94 ms -> 14.07 ms | Improved |
| use PG-SP instead of BFTT3D for memory | PG-SP · GPU memory | MiB: 7400 MiB -> 1300 MiB | Improved |
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.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 3 of 3 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
- SMART-PC: Skeletal Model Adaptation for Robust Test-Time Training in Point Clouds (ICML 2025 · similarity 0.88)
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 2025 · similarity 0.87)
- PointMAC: Meta-Learned Adaptation for Robust Test-Time Point Cloud Completion (NeurIPS 2025 · similarity 0.86)
- Backpropagation-free Network for 3D Test-time Adaptation (CVPR 2024 · similarity 0.86)
- PointLoRA: Low-Rank Adaptation with Token Selection for Point Cloud Learning (CVPR 2025 · similarity 0.85)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · 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.
