Research map / Point clouds / Point cloud classification
Low-Rank Test-Time Training for Pre-Trained Point Cloud Models
Ouyangzi Ye, Feifei Shao, Kexin Li, Yawei Luo, Zikai Song, Ping Liu, Fengda Zhang, Hongwei Wang, Jun Xiao
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 masked feature alignment (MFA) to joint training (ablation) | LoTT-PC · ScanObjectNN-C | accuracy (%) source-only: 48.3 -> 52.0 | Improved |
| use low-rank modulation units instead of full fine-tuning (ablation) | LoTT-PC · ScanObjectNN-C | accuracy (%) online: 51.0 -> 54.1 | Improved |
| online test-time training on top of source-only model | LoTT-PC · ScanObjectNN-C | accuracy (%): 52.0 -> 54.1 | Improved |
| increase augmentation batch size from 1 to 8 masked views (ablation) | LoTT-PC online TTT · ScanObjectNN-C | accuracy (%): 54.1 -> 51.3 | Hurt |
| raise test-time masking ratio from 0.1 to 0.9 (ablation) | LoTT-PC online TTT · ScanObjectNN-C | accuracy (%): 54.1 -> 49.1 | Hurt |
| raise test-time masking ratio from 0.1 to 0.5 (ablation) | LoTT-PC online TTT · ScanObjectNN-C | accuracy (%): 54.1 -> 53.4 | Hurt |
| infer with masked view or ensembles instead of complete view (ablation) | LoTT-PC · ScanObjectNN-C | accuracy (%): 54.1 -> 53.9 (masked), 53.8 (ensemble) | No clear effect |
| replace decoder-based auxiliary objectives with decoder-free MFA plus LoRA units | MATE-Online / SMART-PC-Online · ModelNet40-C | accuracy (%): 71.3 -> 75.5 (vs MATE-Online) | Improved |
| LoTT-PC versus MATE-Online on synthetic ShapeNet-C | LoTT-PC-Online · ShapeNet-C | accuracy (%): 67.4 vs 69.1 | Mixed |
Closest papers by meaning
- SMART-PC: Skeletal Model Adaptation for Robust Test-Time Training in Point Clouds (ICML 2025 · similarity 0.90)
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 2025 · similarity 0.88)
- PointMAC: Meta-Learned Adaptation for Robust Test-Time Point Cloud Completion (NeurIPS 2025 · similarity 0.88)
- PointLoRA: Low-Rank Adaptation with Token Selection for Point Cloud Learning (CVPR 2025 · similarity 0.88)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.87)
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.
