Research map / Vision-language and multimodal / Multimodal LLMs
Make Geometry Matter for Spatial Reasoning
Shihua Zhang, Qiuhong Shen, Shizun Wang, Tianbo Pan, Xinchao Wang
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 Geometry-Guided Fusion instead of original fusion (ablation) | GeoSR · VSI-Bench | Avg accuracy: 50.0 -> 51.9 | Improved |
| use Geometry-Guided Fusion instead of original fusion (ablation) | GeoSR · DSR-Bench | Avg accuracy: 64.7 -> 66.1 | Improved |
| add Geometry-Unleashing Masking to Geo. Fus. (ablation) | GeoSR · DSR-Bench | Avg accuracy: 64.9 -> 66.1 | Improved |
| add Geometry-Unleashing Masking to Geo. Fus. (ablation) | GeoSR · VSI-Bench | Avg accuracy: 50.9 -> 51.9 | Improved |
| masking without any fusion module (ablation) | GeoSR · DSR-Bench | Avg accuracy: 66.1 -> 58.1 | Hurt |
| naive geometry injection (original fusion) vs no geometry (ablation) | Qwen2.5-VL-7B · DSR-Bench | Avg accuracy: 64.0 -> 62.8 | Hurt |
| naive geometry injection (original fusion) vs no geometry (ablation) | Qwen2.5-VL-7B · VSI-Bench | Avg accuracy: 49.8 -> 50.2 | No clear effect |
| full GeoSR vs no geometry modules (ablation) | Qwen2.5-VL-7B · DSR-Bench | Avg accuracy: 64.0 -> 66.1 | Improved |
| full GeoSR vs no geometry modules (ablation) | Qwen2.5-VL-7B · VSI-Bench | Avg accuracy: 49.8 -> 51.9 | Improved |
| lower masking ratio gamma (0.8 to 0.4) (ablation) | GeoSR · DSR-Bench | Avg accuracy: 66.1 -> 65.0 | Hurt |
| higher masking probability beta (0.5 to 0.7) (ablation) | GeoSR · DSR-Bench | Avg accuracy: 66.1 -> 64.5 | Hurt |
| lower masking probability beta (0.5 to 0.3) (ablation) | GeoSR · DSR-Bench | Avg accuracy: 66.1 -> 64.9 | Hurt |
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.
- VLM-3R — VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026) · better on 1 of 1 shared result
- Spatial-MLLM — Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 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
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.92)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.92)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.92)
- Thinking with Geometry: Active Geometry Integration for Spatial Reasoning (ICML 2026 · similarity 0.92)
- Think with 3D: Geometric Imagination Grounded Spatial Reasoning from Limited Views (CVPR 2026 · 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.
