Research map / Large language models / Language, safety and interpretability
3D-LLM: Injecting the 3D World into Large Language Models
Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, Chuang Gan
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 |
|---|---|---|---|
| inject 3D features instead of multi-view image features (ablation) | BLIP2-FlanT5 backbone · ScanQA val | CIDEr: 45.7 -> 69.4 | Improved |
| use 3D features instead of single-image features (ablation) | BLIP2-FlanT5 backbone · ScanQA val | CIDEr: 42.6 -> 69.4 | Improved |
| use 3D-LLM instead of ScanQA with explicit object detection | ScanQA · ScanQA val | BLEU-1: 30.2 -> 39.3 | Improved |
| use 3D-LLM instead of ScanQA | ScanQA · ScanQA test | EM: 23.5 -> 19.1 | Hurt |
| use 3D-LLM (BLIP2-flant5) instead of ScanQA | ScanQA · ScanQA test | CIDEr: 67.3 -> 69.6 | Improved |
| use 3D-LLM instead of 3D-CLR | 3DMV-VQA · 3DMV-VQA | overall accuracy: 57.7 -> 58.6 | Improved |
| use 3D-LLM instead of 3D-CLR | 3DMV-VQA · 3DMV-VQA | counting accuracy: 41.3 -> 32.4 | Hurt |
| use 3D-LLM instead of ScanRefer with VoteNet | ScanRefer · ScanRefer val | ACC@0.25: 41.2 -> 30.3 | Hurt |
| add the Flamingo perceiver resampler (ablation) | 3D-LLM (flamingo) · ScanQA val | CIDEr: 58.9 -> 59.2 | Mixed |
| use 3D-LLM instead of language-only FlanT5 (ablation) | 3D-assisted dialog · held-in 3D-assisted dialog | BLEU-4: 8.7 -> 16.6 | Improved |
| use BLIP2-FlanT5 backbone instead of Flamingo for 3D-LLM (ablation) | 3D-LLM · ScanQA val | CIDEr: 59.2 -> 69.4 | Improved |
Closest papers by meaning
- 3D-PLOT-LLM: Part-Level Object Tokens for 3D Large Language Models (NeurIPS 2026 · similarity 0.93)
- Language-Image Models with 3D Understanding (ICLR 2025 · similarity 0.92)
- Vid-LLM: A Compact Video-based 3D Multimodal LLM with Reconstruction–Reasoning Synergy (ICLR 2026 · similarity 0.92)
- LiDAR-LLM: Exploring the Potential of Large Language Models for 3D LiDAR Understanding (AAAI 2025 · similarity 0.92)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.91)
- Video-3D LLM: Learning Position-Aware Video Representation for 3D Scene Understanding (CVPR 2025 · similarity 0.91)
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.
