Research map / Detection and segmentation / Open-set and few-shot detection
ViTPrompt: Training-Free Prompt Refinement with Visual Tokens for Open-Vocabulary Detection
Yitong Qin, Lihua Zhou, Jiwei Wei, Ran Ran, Shiyuan He, Zeyu Ma, Shuaifeng Li, Nianxin Li, Heng Tao Shen
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 RoI-aligned visual tokens to the text prompt (max fusion) (ablation) | ViTPrompt on Grounding DINO Swin-T · COCO-C (avg) | mAP50: 25.57 -> 26.25 | Improved |
| add RoI-aligned visual tokens to the text prompt (max fusion) (ablation) | ViTPrompt on Grounding DINO Swin-T · FoggyCityscapes | mAP50: 28.89 -> 29.10 | Improved |
| two-stage visual token prompting for localization | Grounding DINO Swin-T · FoggyCityscapes | mAP@0.5:0.95: 15.65 -> 19.02 | Improved |
| two-stage visual token prompting for localization | Grounding DINO Swin-T · COCO-C (avg) | mAP@0.5:0.95: 15.27 -> 18.36 | Improved |
| increase top-K visual tokens beyond 10 (ablation) | ViTPrompt · FoggyCityscapes / COCO-C | mAP50: FoggyCityscapes 29.20 -> 28.76 across K | No clear effect |
| vary NMS IoU threshold between 0.2 and 0.8 (ablation) | ViTPrompt · COCO-C and FoggyCityscapes | mAP50: COCO-C 26.02 -> 26.30 | No clear effect |
| ViTPrompt versus BCA+ on PASCAL-C | ViTPrompt · PASCAL-C Swin-T | mAP50: 40.67 -> 43.03 | Improved |
| ViTPrompt versus BCA+ on COCO-C | ViTPrompt · COCO-C Swin-B | mAP50: 39.98 -> 42.28 | Improved |
| ViTPrompt versus BCA on Rider class | ViTPrompt · FoggyCityscapes Swin-T per-class Rider | AP50: 5.61 -> 4.72 | Hurt |
| add second forward pass with visual tokens | Grounding DINO Swin-T · Efficiency 600x600 | FLOPs (G): 145.64 -> 252.26 | Hurt |
Closest papers by meaning
- Region-Aware Pretraining for Open-Vocabulary Object Detection With Vision Transformers (CVPR 2023 · similarity 0.89)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.88)
- Test-Time Adaptive Object Detection with Foundation Model (NeurIPS 2025 · similarity 0.88)
- LLMs Meet VLMs: Boost Open Vocabulary Object Detection with Fine-grained Descriptors (ICLR 2024 · similarity 0.88)
- DitHub: A Modular Framework for Incremental Open-Vocabulary Object Detection (NeurIPS 2025 · similarity 0.87)
- Exploring Region-Word Alignment in Built-in Detector for Open-Vocabulary Object Detection (CVPR 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.
