Research map / Vision-language and multimodal / Multimodal LLMs
Teaching VLMs to Localize Specific Objects from In-context Examples
Sivan Doveh, Nimrod Shabtay, Eli Schwartz, Hilde Kuehne, Raja Giryes, Rogerio Feris, Leonid Karlinsky, James Glass, Assaf Arbelle, Shimon Ullman, M. Jehanzeb Mirza
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 pseudo-name regularization (replace true class names with pseudo-names) to the fine-tuning data (ablation) | IPLoc 7B · Few-shot personalized localization average | mIoU (%): 30.98 -> 43.31 | Improved |
| add pseudo-name regularization to the fine-tuning data (ablation) | IPLoc 72B · Few-shot personalized localization average | mIoU (%): 47.98 -> 55.79 | Improved |
| fine-tune on all three tracking datasets (LASOT, GOT, TAO) instead of LASOT only (ablation) | IPLoc 7B · ICL-LASOT | mIoU (%): 30.30 -> 55.36 | Improved |
| fine-tune on TAO only versus the full three-dataset mix (ablation) | IPLoc 7B · ICL-LASOT | mIoU (%): 55.36 -> 14.06 | Hurt |
| sample few-shot conversations from the same video instead of arbitrary videos (ablation) | IPLoc 7B · PDM | mIoU (%): 28.94 -> 50.05 | Improved |
| sample few-shot conversations from the same video instead of arbitrary videos (ablation) | IPLoc 7B · ICL-LASOT | mIoU (%): 24.14 -> 39.13 | Improved |
| fine-tune with LoRA on the localization data (generalization retention check) | Qwen2-VL-72B · GQA | accuracy: 64.66 -> 63.37 | Hurt |
| fine-tune with LoRA on the localization data (generalization retention check) | Qwen2-VL-72B · SEED | accuracy: 77.64 -> 77.66 | No clear effect |
| fine-tune with IPLoc data (pseudo-name test, names swapped for pseudo names) | Qwen2-VL-7B · LASOT test set with pseudo names | mIoU (%): 16.55 -> 44.76 | Improved |
| apply the IPLoc fine-tuning recipe to LLaVA-OV | LLaVA-OV-7B · Few-shot personalized localization average | mIoU (%): 20.65 -> 27.25 | Improved |
| scale base model from 7B to 72B | Qwen2-VL · Few-shot personalized localization average | mIoU (%): 31.77 -> 34.21 | Mixed |
| use a more explicit prompt instead of the original training prompt (ablation) | Qwen2-VL-72B · Few-shot personalized localization average | IoU (%): 7.69 -> 34.21 | Improved |
Closest papers by meaning
- Mechanisms of Object Localization in Vision-Language Models (CVPR 2026 · similarity 0.90)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.88)
- Transferable Video Moment Localization by Moment-Guided Query Prompting (AAAI 2024 · similarity 0.88)
- Grounding Everything: Emerging Localization Properties in Vision-Language Transformers (CVPR 2024 · similarity 0.88)
- Learning to Localize Objects Improves Spatial Reasoning in Visual-LLMs (CVPR 2024 · similarity 0.88)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.88)
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.
