Research map / Vision-language and multimodal / Multimodal LLMs
3DZip: Spatial-Aware Feature Diversity-Guided Token Compression for 3D Question Answering
Changwoo Baek, Kyeongbo Kong
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 feature-space distance instead of spatial distance in DPP anchor selection (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 50.1 -> 52.8 | Improved |
| use feature-space distance instead of spatial distance in DPP anchor selection (ablation) | 3DZip · SQA3D test (32 tokens) | EM: 49.0 -> 51.1 | Improved |
| add coarse voxelization before anchor selection (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 51.8 -> 52.8 | Improved |
| increase voxel size from 0.2m to 0.3m (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.8 -> 52.1 | Hurt |
| add spatially-constrained token merging (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.5 -> 52.8 | Improved |
| add spatial constraint to token merging (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.3 -> 52.8 | Improved |
| use DPP instead of farthest point sampling for anchor selection (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.2 -> 52.8 | Improved |
| use DPP instead of MMDP for anchor selection (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.5 -> 52.8 | Improved |
| use cosine similarity instead of L1 distance in feature space (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.5 -> 52.8 | Improved |
| increase merging threshold tau_g from 1 to 5 (ablation) | 3DZip · SQA3D test (64 tokens) | EM: 52.5 -> 52.8 | Improved |
| add DPP anchor selection on top of voxel-mean aggregation (ablation) | Voxel-mean · SQA3D test (64 tokens) | EM: 49.8 -> 52.5 | Improved |
| add DPP anchor selection on top of ConceptFusion aggregation (ablation) | ConceptFusion · SQA3D test (64 tokens) | EM: 50.3 -> 52.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 3 of 3 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
- Zero-shot 3D Question Answering via Voxel-based Dynamic Token Compression (CVPR 2025 · similarity 0.93)
- Merge3D: Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging (CVPR 2026 · similarity 0.89)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.89)
- Proxy3D: Efficient 3D Representations for Vision-Language Models via Semantic Clustering and Alignment (CVPR 2026 · 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 0.88)
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.
