Research map / Vision-language and multimodal / Multimodal LLMs
MonoSR: Open-Vocabulary Spatial Reasoning on Monocular Images
Qirui Wang, Jingyi He, Yining Pan, Si Yong Yeo, Xulei Yang, Shijie Li
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 MonoSR | Qwen2.5-VL-3B · MonoSR Indoor | overall accuracy: 0.283 -> 0.494 | Improved |
| fine-tune on MonoSR | Qwen2.5-VL-3B · MonoSR Outdoor | overall accuracy: 0.302 -> 0.442 | Improved |
| fine-tune on MonoSR | InternVL-3.5-2B · MonoSR Indoor | overall accuracy: 0.301 -> 0.483 | Improved |
| fine-tune on MonoSR | InternVL-3.5-2B · MonoSR Outdoor | overall accuracy: 0.294 -> 0.462 | Improved |
| add scene information (SI) text (ablation) | fine-tuned VLM · MonoSR low/mid-level tasks | accuracy | Improved |
| add 2D visual prompts (ablation) | fine-tuned VLM · MonoSR mid-level indoor | accuracy | Improved |
| add 3D bounding boxes (ablation) | fine-tuned VLM · MonoSR | accuracy | Improved |
| add targeted 2D VP + 3D bbox to fine-tuned model (ablation) | Qwen2.5-VL-3B · MonoSR Indoor | overall accuracy: 0.494 -> 0.798 | Improved |
| provide 2D VP + 3D bbox for all objects instead of targeted (ablation) | Qwen2.5-VL-3B · MonoSR Indoor | overall accuracy: 0.798 -> 0.767 | Hurt |
| provide 2D VP + 3D bbox for all objects instead of targeted (ablation) | InternVL-3.5-2B · MonoSR Indoor | overall accuracy: 0.794 -> 0.721 | Hurt |
| fine-tune on MonoSR | Qwen2.5-VL-3B · MonoSR long-tail classes (bottom 20%) | accuracy: 0.375 -> 0.675 | Improved |
| use stricter 5% relative error threshold | Gemini-2.5-Pro · MonoSR Object-Centric Dist (Num.) | accuracy: 0.305 -> 0.143 | Hurt |
Closest papers by meaning
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.91)
- SSR: Enhancing Depth Perception in Vision-Language Models via Rationale-Guided Spatial Reasoning (NeurIPS 2025 · similarity 0.90)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.90)
- Spatial-DISE: A Unified Benchmark for Evaluating Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.90)
- SpaceVista: All-Scale Visual Spatial Reasoning from mm to km (ICML 2026 · similarity 0.90)
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.
