Research map / Label-efficient and robust learning / Test-time adaptation
Adapt-As-You-Walk Through the Clouds: Training-Free Online Test-Time Adaptation of 3D Vision-Language Foundation Models
Mehran Tamjidi, Hamidreza Dastmalchi, Mohammadreza Alimoradijazi, Ali Cheraghian, Aijun An, Morteza Saberi
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 cluster-based online prototyping to the source model (ablation) | Uni3D-Large · ModelNet-40C | top-1 accuracy: 59.15 -> 68.48 | Improved |
| add Prototype Reassignment (graph label smoothing) on top of online prototyping (ablation) | Uni-Adapter · ModelNet-40C | top-1 accuracy: 68.48 -> 69.70 | Improved |
| add Prototype Reassignment on top of online prototyping (ablation) | Uni-Adapter · ShapeNet-C | top-1 accuracy: 61.79 -> 62.41 | Improved |
| add Prototype Reassignment on top of online prototyping (ablation) | Uni-Adapter · ScanObjectNN-C | top-1 accuracy: 45.12 -> 46.33 | Improved |
| use cluster-based prototypes instead of confidence-based cache (ablation) | Uni-Adapter · ShapeNet-C | top-1 accuracy | Improved |
| set label smoothing lambda_reg near 0 (negligible smoothing) (ablation) | Uni-Adapter · ModelNet-40C | top-1 accuracy | Hurt |
| too many or too few cluster centers (N) versus N=30 (ablation) | Uni-Adapter · ModelNet-40C | top-1 accuracy | Hurt |
| replace high-confidence dual cache (Point-Cache) with cluster cache | Uni3D-Large · ShapeNet-C | top-1 accuracy: 57.70 -> 62.41 | Improved |
| apply Uni-Adapter instead of Point-Cache on Objaverse-LVIS | ULIP-2 · Objaverse-LVIS (clean) | top-1 accuracy: 32.36 -> 31.83 | Hurt |
| add Uni-Adapter to source models vs Point-Cache on clean datasets | Uni3D, OpenShape, ULIP-2 · clean small/large-scale datasets | top-1 accuracy: e.g. ModelNet OpenShape 84.04 -> 85.44 | Mixed |
Closest papers by meaning
- CLIPoint3D: Language-Grounded Few-Shot Unsupervised 3D Point Cloud Domain Adaptation (CVPR 2026 · similarity 0.89)
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 2025 · similarity 0.87)
- Unveiling the Mist over 3D Vision-Language Understanding: Object-centric Evaluation with Chain-of-Analysis (CVPR 2025 · similarity 0.86)
- Point-PEFT: Parameter-Efficient Fine-Tuning for 3D Pre-trained Models (AAAI 2024 · similarity 0.86)
- Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · similarity 0.86)
- MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection (ICLR 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.
