Research map / Vision-language and multimodal / CLIP and zero-shot
pySpatial: Generating 3D Visual Programs for Zero-Shot Spatial Reasoning
Zhanpeng Luo, Ce Zhang, Silong Yong, Cunxi Dai, Qianwei Wang, Haoxi Ran, Guanya Shi, Katia P. Sycara, Yaqi Xie
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 pySpatial 3D visual programming to code agent | GPT-4.1-mini · MindCube (full) | overall accuracy: 45.62 -> 58.56 | Improved |
| add pySpatial (ablation) | GPT-4o · MindCube-1k | overall accuracy: 42.29 -> 62.67 | Improved |
| add pySpatial (ablation) | GPT-4.1 · MindCube-1k | overall accuracy: 44.67 -> 62.35 | Improved |
| add pySpatial (single-view) | GPT-4o · Omni3D-Bench | total score: 42.9 -> 44.2 | Improved |
| add pySpatial | GPT-4o · MMSI-Bench | average accuracy: 30.3 -> 37.3 | Improved |
| use explicit 3D reconstruction tools instead of implicit mental models (CoT, cognitive map) | Qwen2.5-VL-3B mental models vs pySpatial · MindCube-1k | overall accuracy: 41.43 -> 62.35 | Improved |
| give VADAR the 3D reconstruction module (ablation) | VADAR · MindCube-1k | overall accuracy: 40.76 -> 35.62 | Hurt |
| add in-context examples to code agent (0 -> 2) (ablation) | pySpatial GPT-4o · MindCube-1k | overall accuracy: 53.62 -> 62.67 | Improved |
| increase in-context examples from 2 to 4 (ablation) | pySpatial GPT-4o · MindCube-1k | overall accuracy: 62.67 -> 63.14 | No clear effect |
| swap 3D reconstruction backbone (VGGT, Pi3, CUT3R) (ablation) | pySpatial GPT-4o · MindCube-1k | overall accuracy: 62.67 / 63.33 / 61.05 | No clear effect |
| swap code agent (GPT-4o, Qwen3-Coder, DeepSeek-v3) (ablation) | pySpatial · MindCube-1k | overall accuracy: 62.67 / 62.10 / 61.05 | No clear effect |
| use pySpatial on Rotation subset with GPT-4.1-mini agent vs Claude-4-Sonnet | pySpatial · MindCube (full) Rotation | accuracy: 48.42 (Claude-4-Sonnet) vs 43.20 | Hurt |
Later papers that compared with this method
2 later accepted papers report pySpatial as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning (NeurIPS 2026 · better on 1 of 1 shared result)
- From Views to Worlds: Active Exploration over 3D Worlds for Vision-Language Models (NeurIPS 2026 · better on 1 of 1 shared result)
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.
- VLM-3R — VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026) · 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
- RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics (CVPR 2025 · similarity 0.89)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Struct2D: A Perception-Guided Framework for Spatial Reasoning in MLLMs (NeurIPS 2025 · similarity 0.88)
- HiSpatial: Taming Hierarchical 3D Spatial Understanding in Vision-Language Models (CVPR 2026 · similarity 0.88)
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.87)
- Socratic-Geo: Synthetic Data Generation and Cross-Modal Geometric Reasoning via Multi-Agent Interaction (CVPR 2026 · similarity 0.87)
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.
