Research map / Point clouds / Point cloud classification
4DPC$^2$hat: Towards Dynamic Point Cloud Understanding with Failure-Aware Bootstrapping
Xindan Zhang, Weilong Yan, YUFEI SHI, Xuerui Qiu, Tao He, Ying Li, Ming Li, Hehe Fan
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 |
|---|---|---|---|
| replace temporal Transformer with bidirectional Mamba (ablation) | 4DPC2hat · 4D object captioning (Objaverse) | GPT-4 score: 69.08 -> 73.27 | Improved |
| replace temporal Transformer with bidirectional Mamba (ablation) | 4DPC2hat · 4D object QA (Objaverse) | GPT-4 average score: 74.41 -> 78.01 | Improved |
| add two rounds of failure-aware bootstrapping after SFT (ablation) | 4DPC2hat · 4D object QA (Objaverse) | GPT-4 average score: 74.40 -> 78.01 | Improved |
| add failure-aware bootstrapping, temporal category (ablation) | 4DPC2hat · 4D object QA (Objaverse) | SimCSE temporal reasoning: 71.41 -> 76.52 | Improved |
| add failure-aware bootstrapping, counting category (ablation) | 4DPC2hat · 4D object QA (Objaverse) | SimCSE counting: 73.19 -> 77.03 | Improved |
| naive data augmentation with failure-targeted QA instead of bootstrapping (ablation) | 4DPC2hat · 4D object QA (Objaverse) | GPT-4 average score: 74.40 -> 75.87 | Mixed |
| second bootstrapping round beyond the first (ablation) | 4DPC2hat · 4D object QA (Objaverse) | score | Mixed |
| increase bootstrapping data from 12k to 15k samples (ablation) | 4DPC2hat · 4D object QA (Objaverse) | score | No clear effect |
| process frames independently with static 3D MLLMs plus temporal aggregation instead of sequence modeling | PointLLM, ShapeLLM, MiniGPT-3D · 4D object captioning (Objaverse) | GPT-4 score: 54.70 -> 73.27 | Hurt |
Closest papers by meaning
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 2025 · similarity 0.88)
- PointRWKV: Efficient RWKV-Like Model for Hierarchical Point Cloud Learning (AAAI 2025 · similarity 0.88)
- More Text, Less Point: Towards 3D Data-Efficient Point-Language Understanding (AAAI 2025 · similarity 0.88)
- X4D-SceneFormer: Enhanced Scene Understanding on 4D Point Cloud Videos through Cross-Modal Knowledge Transfer (AAAI 2024 · similarity 0.87)
- Learning to Reason in 4D: Dynamic Spatial Understanding for Vision Language Models (CVPR 2026 · similarity 0.87)
- Adapting Pre-trained 3D Models for Point Cloud Video Understanding via Cross-frame Spatio-temporal Perception (CVPR 2025 · 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.
