Research map / Vision-language and multimodal / Human and embodied modeling
Visual Grounding for Object Questions
Martin Nicolas Everaert, Xiruo Liu, Hiroyuki Takeda, Raja Bala, Vivek Yadav, Vidya Narayanan
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 |
|---|---|---|---|
| rephrase visual question as general object question | SoTA grounding models (UnifiedIO-XL) · VizWiz images, specific visual evidence subset | gIoU (%): 52.2 -> 37.2 | Hurt |
| train lightweight CLIPSeg-style model on VGOQ data jointly with RES and VQA grounding data | Our LW (1.77M parameters) · ABO-VGOQ val specific visual evidence | gIoU (%): 12.4 (UnifiedIO-XL) -> 39.5 | Improved |
| compare 1.77M lightweight model with Qwen3-VL-8B | Our LW · ABO-VGOQ test specific visual evidence | gIoU (%): 30.3 -> 32.5 (Qwen3-VL-4B 32.7) | No clear effect |
| compare lightweight model with Qwen3-VL-8B on VizWiz-VGOQ related-to-question | Our LW · VizWiz-VGOQ val-RTQ | gIoU (%): 49.4 -> 64.1 | Improved |
| compare lightweight model on Toloka VQA grounding (not trained on) | Our LW · Toloka test-public | gIoU (%): 71.4 (Qwen3-VL-8B) -> 12.6 | Hurt |
| lightweight model versus UnifiedIO-XL on VizWiz-VQA-Grounding VQ | Our LW · VizWiz-VQA-Grounding val | gIoU (%): 68.1 -> 67.2 | Hurt |
| finetune only on specific-visual-evidence samples (ablation) | Our LW · ABO-VGOQ / VizWiz-VGOQ specific evidence | gIoU (percentage points): +1.7 to +7.4 | Improved |
| scale UnifiedIO from Small to XL | UnifiedIO · ABO-VGOQ val specific visual evidence | gIoU (%): 12.3 -> 12.4 | No clear effect |
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.
- UnifiedIO-XL — UNIFIED-IO: A Unified Model for Vision, Language, and Multi-modal Tasks (ICLR 2023) · better on 6 of 8 shared results
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
- Visual Grounding via Accumulated Attention (CVPR 2018 · similarity 0.90)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.87)
- Can I Trust Your Answer? Visually Grounded Video Question Answering (CVPR 2024 · similarity 0.87)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.87)
- ConVQG: Contrastive Visual Question Generation with Multimodal Guidance (AAAI 2024 · similarity 0.87)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · 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.
