Research map / Vision-language and multimodal / Visual question answering
Weakly Supervised Relative Spatial Reasoning for Visual Question Answering
Pratyay Banerjee, Tejas Gokhale, Yezhou Yang, Chitta Baral
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 weak supervision from 3D relative position estimation (15-bin classification) (ablation) | LXMERT + SR · GQA-Val | accuracy: 59.40 -> 62.32 | Improved |
| add weak supervision from 2D object centroid estimation (regression) (ablation) | LXMERT + SR · GQA-Val | accuracy: 59.40 -> 57.33 | Hurt |
| add weak supervision from 3D object centroid estimation (regression) (ablation) | LXMERT + SR · GQA-Val | accuracy: 59.40 -> 58.28 | Hurt |
| add weak supervision from 2D object centroid estimation (15-bin classification) (ablation) | LXMERT + SR · GQA-Val | accuracy: 59.40 -> 58.64 | Hurt |
| add weak supervision from 2D relative position regression (ablation) | LXMERT + SR · GQA-Val | accuracy: 59.40 -> 59.85 | Improved |
| use early fusion of relative position vectors instead of late fusion (ablation) | LXMERT + SR · 3D RPE 15-way | accuracy: 69.45 -> 74.20 | Improved |
| add relative position vectors via early fusion (ablation) | LXMERT + SR · 3D RPE 15-way | accuracy: 52.80 -> 74.20 | Improved |
| add spatial pyramid image patches to early fusion (ablation) | LXMERT + SR + Early Fusion · 3D RPE 15-way | accuracy: 74.20 -> 81.47 | Improved |
| add spatial pyramid image patches to late fusion (ablation) | LXMERT + SR + Late Fusion · 3D RPE 15-way | accuracy: 69.45 -> 79.60 | Improved |
| add 9x9 grid patches to the patch set (ablation) | Spatial pyramid patches · GQA-Val | accuracy | No clear effect |
| extract patch features with ResNet101 instead of ResNet-50 (ablation) | Spatial pyramid patches · GQA-Val | accuracy: +0.05% | Improved |
| add weakly supervised relative position estimation with early fusion and patches | LXMERT · GQA test-standard | accuracy: 60.34 -> 62.11 | Improved |
Closest papers by meaning
- SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities (CVPR 2024 · similarity 0.92)
- SpatiaLQA: A Benchmark for Evaluating Spatial Logical Reasoning in Vision-Language Models (CVPR 2026 · similarity 0.92)
- Enhancing Spatial Reasoning Through Visual and Textual Thinking (AAAI 2026 · similarity 0.91)
- Pursuing Minimal Sufficiency in Spatial Reasoning (ICLR 2026 · similarity 0.91)
- SpatialReasoner: Towards Explicit and Generalizable 3D Spatial Reasoning (NeurIPS 2025 · similarity 0.91)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 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.
