Research map / Point clouds / Point cloud classification
MATE: Masked Autoencoders are Online 3D Test-Time Learners
M. Jehanzeb Mirza, Inkyu Shin, Wei Lin, Andreas Schriebl, Kunyang Sun, Jaesung Choe, Mateusz Kozinski, Horst Possegger, In So Kweon, Kuk-Jin Yoon, Horst Bischof
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-autoencoder test-time training (MATE-Online) on top of joint-trained network | PointMAE · ModelNet-40C | Top-1 accuracy (%): 53.9 -> 71.3 | Improved |
| apply entropy-minimization TTT (TENT, SHOT) at batch size 1 | PointMAE · ModelNet-40C | Top-1 accuracy (%): 53.9 -> 26.5 (TENT), 26.6 (SHOT) | Hurt |
| use rotation prediction as the TTT auxiliary task instead of MAE reconstruction (ablation) | PointMAE · ModelNet-40C | Top-1 accuracy (%): 53.9 -> 53.0 (TTT-Rot) | Hurt |
| use larger test-time batch size (128) for baselines (ablation) | TENT, SHOT, T3A · ModelNet-40C | Top-1 accuracy (%): TENT 26.5 -> 65.6, SHOT 26.6 -> 63.8 | Improved |
| increase masking ratio from 60% to 95% (ablation) | MATE-Online · ModelNet-40C | Top-1 accuracy (%): 71.5 -> 71.6 | No clear effect |
| increase masking ratio to 97.5% (ablation) | MATE-Online · ModelNet-40C | Top-1 accuracy (%): 71.6 -> 56.9 | Hurt |
| increase TTT batch size from 1 to 2 (ablation) | MATE-Online · ModelNet-40C | Top-1 accuracy (%): 43.1 -> 66.4 | Improved |
| increase TTT batch size from 8 to 48 (ablation) | MATE-Online · ModelNet-40C | Top-1 accuracy (%): 69.7 -> 71.3 | Improved |
| adapt only every 5th sample (stride 5) (ablation) | MATE · ShapeNet-C | Top-1 accuracy (%): about 3 percent-point drop, 20 FPS | Hurt |
| adapt only every 300th sample (stride 300) (ablation) | MATE · ShapeNet-C | Top-1 accuracy (%): ~3 percent-points above Source-Only at 62 FPS | Improved |
| use MATE-Standard (reset per sample) vs MATE-Online (accumulate) (ablation) | MATE · ModelNet-40C | Top-1 accuracy (%): 64.3 -> 71.3 | Improved |
| apply MATE to combined corruptions | PointMAE · ModelNet-40 combined corruptions | Top-1 accuracy (%): 30.6 -> 53.2 | Improved |
Closest papers by meaning
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.89)
- Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training (NeurIPS 2022 · similarity 0.88)
- Point-MaDi: Masked Autoencoding with Diffusion for Point Cloud Pre-training (NeurIPS 2025 · similarity 0.88)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.87)
- SMART-PC: Skeletal Model Adaptation for Robust Test-Time Training in Point Clouds (ICML 2025 · similarity 0.87)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.86)
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.
