Research map / Vision-language and multimodal / Human and embodied modeling
Improving Visual Grounding With Visual-Linguistic Verification and Iterative Reasoning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 VLTVG instead of TransVG | RefCOCO, RefCOCO+, RefCOCOg, ReferItGame, and Flickr30k Entities visual grounding · RefCOCO, RefCOCO+, RefCOCOg, ReferItGame, Flickr30k Entities | accuracy: 84.77 vs 81.02 (RefCOCO val, ResNet-101); 74.19 vs 64.82 (RefCOCO+ val, ResNet-101); 72.98 vs 67.02 (RefCOCOg val-g, Res | Helped |
| Add multi-stage cross-modal reasoning instead of a single-stage decoder (ablation) | VLTVG · RefCOCOg val-g | accuracy: 63.64 -> 66.02 | Helped |
| Add the language-guided context encoder (ablation) | VLTVG with multi-stage decoder · RefCOCOg val-g | accuracy: 66.02 -> 68.44 | Helped |
| Add visual-linguistic verification (ablation) | VLTVG with multi-stage decoder and context encoder · RefCOCOg val-g | accuracy: 68.44 -> 71.62 (+3.18 percentage points) | Helped |
| Increase decoder stages from 1 to 2 (ablation) | VLTVG decoder · RefCOCOg val-g | accuracy: 63.64 -> 65.05 | Helped |
| Increase decoder stages from 2 to 4 (ablation) | VLTVG decoder · RefCOCOg val-g | accuracy: 65.05 -> 65.70 | Helped |
| Increase decoder stages from 4 to 6 (ablation) | VLTVG decoder · RefCOCOg val-g | accuracy: 65.70 -> 66.02 | Helped |
| Increase decoder stages from 6 to 8 (ablation) | VLTVG decoder · RefCOCOg val-g | accuracy: 66.02 -> 65.97 | Hurt |
| Replace verification and context modules with one transformer encoder layer (ablation) | VLTVG feature learning · RefCOCOg val-g | accuracy: 66.02 -> 69.37 | Helped |
| Replace verification and context modules with two transformer encoder layers (ablation) | VLTVG feature learning · RefCOCOg val-g | accuracy: 69.37 -> 69.22 | No clear effect |
| Replace verification and context modules with three transformer encoder layers (ablation) | VLTVG feature learning · RefCOCOg val-g | accuracy: 69.37 -> 69.15 | No clear effect |
| Replace verification and context modules with four transformer encoder layers (ablation) | VLTVG feature learning · RefCOCOg val-g | accuracy: 69.37 -> 69.55 | No clear effect |
Closest papers by meaning
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · similarity 0.92)
- Look Around Before Locating: Considering Content and Structure Information for Visual Grounding (AAAI 2025 · similarity 0.92)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.92)
- Connecting the Dots: Training-Free Visual Grounding via Agentic Reasoning (AAAI 2026 · similarity 0.91)
- VGent: Visual Grounding via Modular Design for Disentangling Reasoning and Prediction (CVPR 2026 · similarity 0.91)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (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.