Research map / Video / Video-language understanding
Memory Tree Guided Key Frame Querying for Efficient 3D Question Answering
Hsiang-Wei Huang, Fu-Chen Chen, Li-Wu Tsao, Cheng-Han Lee, Che-Chun Su, Lu Xia, Ronghui Peng, Jenq-Neng Hwang, Min Sun, Cheng-Hao Kuo
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 |
|---|---|---|---|
| use MemTree3D key frame selection instead of uniform sampling | GPT-4o · OpenEQA | LLM-Match: 49.4 -> 66.8 | Improved |
| use MemTree3D key frame selection instead of uniform sampling | LLaVA-OneVision-7B · OpenEQA | LLM-Match: 49.2 -> 55.0 | Improved |
| use MemTree3D instead of uniform sampling | LLaVA-OneVision-7B · ScanQA | EM@1: 25.1 -> 28.0 | Improved |
| use MemTree3D instead of uniform sampling | LLaVA-OneVision-7B · SQA3D | EM@1: 46.2 -> 49.6 | Improved |
| add MemTree3D on larger HM3D scenes versus smaller ScanNet scenes | GPT-4o · OpenEQA HM3D vs ScanNet | LLM-Match: +24.6 vs +13.6 | Improved |
| use 6-DoF pose segmentation instead of uniform 90-frame segmentation (ablation) | MemTree3D LocNode · OpenEQA | LLM-Match: 33.0 -> 49.1 | Improved |
| use 6-DoF pose segmentation instead of uniform 30-frame segmentation (ablation) | MemTree3D LocNode · OpenEQA | LLM-Match: 40.4 -> 49.1 | Improved |
| shorten uniform segmentation window from 90 to 30 frames (ablation) | MemTree3D LocNode · OpenEQA | LLM-Match: 33.0 -> 40.4 | Improved |
| use GPT-4o instead of Qwen3-4B as LLM (ablation) | MemTree3D with LLaVA-OneVision-7B · OpenEQA | LLM-Match: 51.1 -> 55.0 | Improved |
| use Qwen3-8B instead of Qwen3-4B as LLM (ablation) | MemTree3D with LLaVA-OneVision-7B · OpenEQA | LLM-Match: 51.1 -> 53.1 | Improved |
| use GPT-4o VLM instead of LLaVA-OneVision-7B (ablation) | MemTree3D with GPT-4o LLM · OpenEQA | LLM-Match: 55.0 -> 66.8 | Improved |
| query objects absent from scene representation (perception failure) | MemTree3D · OpenEQA | LLM-Match: 69.0 -> 65.9 | Hurt |
Closest papers by meaning
- Zero-shot 3D Question Answering via Voxel-based Dynamic Token Compression (CVPR 2025 · similarity 0.89)
- 3D Question Answering via only 2D Vision-Language Models (ICML 2025 · similarity 0.88)
- Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA (AAAI 2024 · similarity 0.87)
- Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering (ICLR 2025 · similarity 0.87)
- Hypo3D: Exploring Hypothetical Reasoning in 3D (ICML 2025 · similarity 0.87)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · 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.
