Research map / Detection and segmentation / Open-set and few-shot detection
DetPO: In-Context Learning with Multi-Modal LLMs for Few-Shot Object Detection
Gautam Rajendrakumar Gare, Neehar Peri, Matvei Popov, Shruti Jain, John Galeotti, Deva Ramanan
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 few-shot images as in-context examples to class names and instructions (ablation) | Qwen3-VL (8B) · RF20-VL (20 datasets, 10-shot) | mAP: 11.4 -> 7.0 | Hurt |
| add few-shot images as in-context examples (ablation) | Qwen2.5-VL (7B) · RF20-VL (20 datasets, 10-shot) | mAP: 6.2 -> 1.8 | Hurt |
| add few-shot images as in-context examples (ablation) | Gemini 3 Pro · RF20-VL (20 datasets, 10-shot) | mAP: 23.0 -> 23.9 | Improved |
| add class instructions to class names (ablation) | Qwen3-VL (30B-A3B) · RF20-VL (20 datasets, 10-shot) | mAP: 10.7 -> 11.9 | Improved |
| apply DetPO prompt optimization | Qwen3-VL (30B-A3B) · RF20-VL (20 datasets, 10-shot) | mAP: 11.9 -> 19.4 | Improved |
| add VQA Score confidence re-scoring (ablation) | Qwen3-VL (30B-A3B) + DetPO · RF20-VL (20 datasets, 10-shot) | mAP: 19.4 -> 21.6 | Improved |
| replace self-reported confidence with SigLIPv2 score (ablation) | Qwen3-VL (30B-A3B) + DetPO · RF20-VL (20 datasets, 10-shot) | mAP: 19.4 -> 16.4 | Hurt |
| use DetPO instead of GEPA | Gemini 3 Pro · RF20-VL (20 datasets, 10-shot) | mAP: 25.6 -> 26.4 | Improved |
| use MIPROv2 prompt optimization | Qwen3-VL (30B-A3B) · RF20-VL (20 datasets, 10-shot) | mAP: 11.9 -> 10.7 | Hurt |
| apply DetPO prompt optimization | Qwen2.5-VL (7B) · RF20-VL (20 datasets, 10-shot) | mAP: 6.2 -> 9.1 | Improved |
| apply DetPO prompt optimization | Qwen2.5-VL (72B) · RF20-VL (20 datasets, 10-shot) | mAP: 10.4 -> 15.7 | Improved |
| add VQA Score re-scoring (ablation) | Qwen3-VL (8B) + DetPO · RF20-VL (20 datasets, 10-shot) | mAP: 15.3 -> 17.5 | Improved |
Closest papers by meaning
- Multi-Modal Few-Shot Object Detection with Meta-Learning-Based Cross-Modal Prompting (IJCV 2026 · similarity 0.91)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.89)
- Multi-Perspective Data Augmentation for Few-shot Object Detection (ICLR 2025 · similarity 0.88)
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 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.
