Research map / Vision-language and multimodal / Multimodal LLMs
OpenSpatial: A Principled Data Engine for Empowering Spatial Intelligence
Jianhui Liu, Haoze Sun, Wenbo Li, Yanbing Zhang, Rui Yang, zhiliang zhu, Yijun Yang, Shenghe Zheng, Nan Jiang, Jiaxiu Jiang, Haoyang Huang, Tien-Tsin Wong and 2 more
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 |
|---|---|---|---|
| fine-tune on OpenSpatial-3M | Qwen2.5-VL-7B · 3D-Avg | accuracy: 50.0 -> 59.5 | Improved |
| fine-tune on OpenSpatial-3M | InternVL2.5-8B · 3D-Avg | accuracy: 51.6 -> 59.3 | Improved |
| fine-tune on OpenSpatial-3M | InternVL3-8B · 3D-Avg | accuracy: 53.2 -> 59.8 | Improved |
| fine-tune on OpenSpatial-3M | Qwen3-VL-8B · 3D-Avg | accuracy: 56.7 -> 62.1 | Improved |
| fine-tune on OpenSpatial-3M | InternVL3-8B · RealWorldQA | accuracy: 70.6 -> 66.6 | Hurt |
| use 3D-box-centric representation instead of point-cloud-centric (ablation) | data engine · BLINK | accuracy: 57.2 -> 60.3 | Improved |
| add occlusion filtering to 3D-box-centric data (ablation) | data engine · CV-3D | accuracy: 78.2 -> 89.9 | Improved |
| add occlusion filtering to 3D-box-centric data (ablation) | data engine · VSI | accuracy: 32.1 -> 41.7 | Improved |
| increase data from 20% to 100% | Qwen2.5-VL-7B · 3D-Avg | accuracy: 57.6 -> 59.7 | Improved |
| scale model from 3B to 32B | OpenSpatial Qwen2.5-VL · 3D-Avg | accuracy: 56.1 -> 61.3 | Improved |
| scale model from 7B to 32B | OpenSpatial Qwen2.5-VL · CV-3D | accuracy: 92.5 -> 93.4 | Hurt |
Closest papers by meaning
- Scaling Spatial Intelligence with Multimodal Foundation Models (CVPR 2026 · similarity 0.91)
- SpatialScore: Towards Comprehensive Evaluation for Spatial Intelligence (CVPR 2026 · similarity 0.90)
- SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning (NeurIPS 2026 · similarity 0.90)
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.90)
- MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence (ICLR 2026 · similarity 0.89)
- Spatial-IQ: Deconstructing Spatial Intelligence via Hierarchical Capability Tests (NeurIPS 2026 · similarity 0.89)
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.
