Research map / Detection and segmentation / Open-set and few-shot detection
Zero-Shot Quantization for Object Detectors using Off-the-Shelf Generative Models
Hyunho Lee, Kyomin Hwang, Hyeonjin Kim, Suyoung Kim, Sunghyun Wee, Nojun Kwak
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 Information-Dense Prompting to diffusion generation (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 16.0 / 28.8 -> 19.3 / 33.8 | Improved |
| add Information-Dense Prompting to diffusion generation (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W8A8 | mAP / mAP50: 33.5 / 52.1 -> 34.3 / 53.5 | Improved |
| add Intrinsic Distribution-Aware Selection to diffusion generation (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 16.0 / 28.8 -> 19.8 / 34.4 | Improved |
| combine IDP and IDAS (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 19.3 / 33.8 -> 21.0 / 36.1 | Improved |
| replace one-hot pseudo-label detect loss with TANR (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W8A8 | mAP / mAP50: 33.9 / 52.8 -> 35.0 / 53.6 | Improved |
| replace one-hot pseudo-label detect loss with TANR (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 20.6 / 35.8 -> 21.0 / 36.1 | Improved |
| increase image pool from 2k to 160k (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W3A3 | mAP / mAP50: 7.76 / 15.2 -> 10.20 / 19.7 | Improved |
| increase image pool from 2k to 160k (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W8A8 | mAP / mAP50: 34.3 / 53.5 -> 35.0 / 53.6 | No clear effect |
| select images by IDAS instead of random (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 19.3 / 33.8 -> 21.0 / 36.1 | Improved |
| select images with most bounding boxes instead of random (ablation) | GoodQ (YOLOv5-s) · MS-COCO val W4A4 | mAP / mAP50: 19.3 / 33.8 -> 17.5 / 30.7 | Hurt |
| use generated diffusion data instead of noise-optimized data (TSOD) at low bit | YOLOv11-s QAT · MS-COCO val W3A3 | mAP: 0.5 -> 14.2 | Improved |
| use GoodQ instead of TSOD on Mask R-CNN Swin-T | Mask R-CNN Swin-T · MS-COCO val W3A3 | mAP / mAP50: 16.5 / 29.5 -> 21.6 / 36.7 | Improved |
Closest papers by meaning
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.87)
- SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning (ICLR 2025 · similarity 0.87)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.87)
- Zero-shot Generalizable Incremental Learning for Vision-Language Object Detection (NeurIPS 2024 · similarity 0.87)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.87)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.86)
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.
