Research map / Vision-language and multimodal / Multimodal LLMs
Contrastive Region Guidance: Improving Grounding in Vision-Language Models without Training
David Wan, Jaemin Cho, Elias Stengel-Eskin, Mohit Bansal
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 contrastive region guidance (blackout contrast) at decoding | LLaVA-1.6-34B · ViP-Bench | average score: 45.0 -> 56.1 | Improved |
| add CRG on top of a visual-prompt fine-tuned model | ViP-LLaVA-13B · ViP-Bench | average score: 48.6 -> 49.4 | Improved |
| add Set-of-Mark overlay prompting to open-source VLMs | LLaVA-1.6-34B and ViP-LLaVA-13B · ViP-Bench | average score: 45.0 -> 42.0 | Hurt |
| overlay bounding boxes without CRG (ablation) | LLaVA-1.6-34B · What'sUp Set of 4 | accuracy: 54.0 -> 39.7 | Hurt |
| add CRG | LLaVA-1.6-34B · What'sUp Set of 4 | accuracy: 54.0 -> 64.0 | Improved |
| add CRG | LLaVA-1.6-34B · SugarCrepe swap-obj | accuracy: 76.3 -> 87.8 | Improved |
| add CRG | LLaVA-1.6-34B · SeeTRUE average | AUROC: 65.7 -> 74.1 | Improved |
| add CRG re-ranking of detector proposals | GroundingDINO-B + LLaVA-1.5-13B · RefCOCO testA (multiple proposals only) | accuracy@0.5: 62.0 -> 66.9 | Improved |
| add CRG re-ranking | GroundingDINO-B + LLaVA-1.5-13B · RefCOCO+ testB (full) | accuracy@0.5: 60.4 -> 60.0 | No clear effect |
| black out each object separately instead of combined mask (ablation) | CRG · What'sUp Set of 4, LLaVA-1.6-34B | accuracy: 60.1 -> 64.0 | Improved |
| black out entire image instead of only relevant objects (ablation) | CRG · What'sUp Set of 4, LLaVA-1.6-34B | accuracy: 64.0 -> 53.8 | Hurt |
| use segmentation-mask blackout instead of bounding-box blackout (ablation) | CRG · What'sUp Set of 4, LLaVA-1.6-34B | accuracy: 64.0 -> 55.7 | Hurt |
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.
- ViP-LLaVA-13B — ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts (CVPR 2024) · better on 2 of 2 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
- RegionGPT: Towards Region Understanding Vision Language Model (CVPR 2024 · similarity 0.91)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.90)
- RegionReasoner: Region-Grounded Multi-Round Visual Reasoning (ICLR 2026 · similarity 0.90)
- Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding (CVPR 2024 · similarity 0.90)
- SIFThinker: Spatially-Aware Image Focus for Visual Reasoning (AAAI 2026 · similarity 0.90)
- VLM-R³: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought (NeurIPS 2025 · 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.
