Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Prompt Tuning for Text-Driven Object Detection
Weizhen He, Weijie Chen, Binbin Chen, Shicai Yang, Di Xie, Luojun Lin, Donglian Qi, Yueting Zhuang
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 Dual Complementary Teaching with offline and online teachers (ablation) | UPT · Cityscapes | mAP50: 40.0 -> 42.4 | Improved |
| add Nested Mean Teaching on top of Dual Complementary Teaching (ablation) | UPT · Cityscapes | mAP50: 42.4 -> 43.1 | Improved |
| feed strongly augmented images to online teacher (ablation) | Dual Complementary Teaching · Cityscapes | mAP50: 40.5 -> 42.4 | Improved |
| use strong instead of weak augmentation for offline teacher (ablation) | Dual Complementary Teaching · Cityscapes | mAP50: 42.4 -> 41.7 | Hurt |
| increase inner-loop steps K from 0 to 3 (ablation) | Nested Mean Teaching · Cityscapes | mAP50: 42.4 -> 43.1 | Improved |
| increase inner-loop steps K from 3 to 4 (ablation) | Nested Mean Teaching · Cityscapes | mAP50: 43.1 -> 42.7 | Hurt |
| tune entire model instead of text prompt without labels (ablation) | GLIP-T unsupervised tuning · Cityscapes | mAP50: 33.8 -> 12.8 | Hurt |
| tune only visual encoder instead of text prompt without labels (ablation) | GLIP-T unsupervised tuning · Cityscapes | mAP50: 33.8 -> 26.5 | Hurt |
| unsupervised prompt tuning vs conventional Unbiased Teacher mean teaching | GLIP-T · Cityscapes | mAP50: 40.0 -> 43.1 | Improved |
| apply UPT to datasets with small zero-shot-to-oracle gap | GLIP-T · MS-COCO | mAP50: 63.5 -> 64.1 | No clear effect |
| apply Unbiased Teacher on single-class datasets | GLIP-T · KITTI/Sim10K/EgoHands/Pistols average | mAP50: 68.5 -> 67.6 | Hurt |
| UPT without labels vs supervised few-shot prompt tuning | GLIP-T · Cityscapes | mAP50: 43.1 vs 42.5 (5-shot) | Improved |
Closest papers by meaning
- DiPEx: Dispersing Prompt Expansion for Class-Agnostic Object Detection (NeurIPS 2024 · similarity 0.88)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.88)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.88)
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 2025 · similarity 0.87)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.87)
- Open-Vocabulary Object Detection via Language Hierarchy (NeurIPS 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.
