Research map / Vision-language and multimodal / CLIP and zero-shot
Zero-Shot 3D Question Answering via Hierarchical View-to-Token Transportation
Dongsheng Wang, Dawei Su, Hui Huang
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 geometry-aware design to key-view selection (ablation) | KeyVT · ScanQA val | CIDEr: 93.8 -> 100.7 | Improved |
| add sub-scene partition (ablation) | KeyVT · ScanQA val | CIDEr: 99.8 -> 100.7 | Improved |
| add relevance scoring (ablation) | KeyVT · ScanQA val | CIDEr: 99.4 -> 100.7 | Improved |
| replace AKS with KeyV for view selection (16 frames) (ablation) | Qwen2.5-VL-7B · VSI-Bench / ScanQA / SQA3D | Acc / CIDEr / EM@1: 36.4 -> 37.7 (VSI), 67.3 -> 70.2 (CIDEr) | Improved |
| add KeyT token compression (16 -> 8 equivalent frames) on top of KeyV (ablation) | LLaVA-Video-7B · VSI-Bench | average accuracy: 33.2 -> 34.6 | Improved |
| add KeyT to uniform sampling (ablation) | Uniform sampling · ScanQA val | CIDEr: 93.4 -> 94.0 | Improved |
| add KeyT to AKS (ablation) | AKS · ScanQA val | CIDEr: 99.0 -> 99.8 | Improved |
| inject 10% noise into camera parameters (ablation) | KeyVT · ScanQA val | CIDEr: 100.7 -> 98.4 | Hurt |
| inject 1-5% noise into camera parameters (ablation) | KeyVT · ScanQA val | CIDEr: 100.7 -> 100.0 (5% noise) | No clear effect |
| use VGGT-estimated cameras instead of true cameras (ablation) | KeyVT · ScanQA val | CIDEr: 100.7 -> 100.0 | No clear effect |
| compress key-view tokens with KeyT instead of DivPrune/FLoC at 50% retain (ablation) | Qwen2.5-VL-7B · ScanQA val | CIDEr: 67.3 / 68.5 -> 69.6 | Improved |
| compress tokens to 50% with KeyT versus full 16 frames (ablation) | Qwen2.5-VL-7B · ScanQA val | CIDEr: 70.2 -> 69.6 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DivPrune — DivPrune: Diversity-based Visual Token Pruning for Large Multimodal Models (CVPR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 3D Question Answering via only 2D Vision-Language Models (ICML 2025 · similarity 0.91)
- Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA (AAAI 2024 · similarity 0.90)
- Zero-shot 3D Question Answering via Voxel-based Dynamic Token Compression (CVPR 2025 · similarity 0.89)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.89)
- Proxy3D: Efficient 3D Representations for Vision-Language Models via Semantic Clustering and Alignment (CVPR 2026 · similarity 0.89)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.88)
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.
