Research map / Detection and segmentation / Weakly supervised detection and segmentation
DAP: Detection-Aware Pre-Training With Weak Supervision
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 CAM pseudo-box generation and detector pre-training after ImageNet-1M classification pre-training | Faster R-CNN FPN ResNet-50 · COCO 2017 val | AP .5:.95: 36.73 -> 37.25 (+0.52) | Helped |
| Add CAM pseudo-box generation and detector pre-training after ImageNet-14M classification pre-training | Faster R-CNN FPN ResNet-50 · COCO 2017 val | AP .5:.95: 38.87 -> 39.57 (+0.70) | Helped |
| Add CAM pseudo-box generation and detector pre-training after ImageNet-1M classification pre-training | Faster R-CNN FPN ResNet-50 · VOC 2007 test, 07 trainval | AP .5: 77.36 -> 79.93 (+2.57) | Helped |
| Add CAM pseudo-box generation and detector pre-training after ImageNet-14M classification pre-training | Faster R-CNN FPN ResNet-50 · VOC 2007 test, 07 trainval | AP .5: 80.74 -> 84.24 (+3.50) | Helped |
| Add CAM pseudo-box generation and detector pre-training after ImageNet-1M classification pre-training | Faster R-CNN FPN ResNet-50 · VOC 2007 test, 07+12 trainval | AP .5: 83.77 -> 84.49 (+0.72) | Helped |
| Add CAM pseudo-box generation and detector pre-training after ImageNet-14M classification pre-training | Faster R-CNN FPN ResNet-50 · VOC 2007 test, 07+12 trainval | AP .5: 86.91 -> 87.84 (+0.93) | Helped |
| Add DAP pre-training rather than classification pre-training | Faster R-CNN FPN ResNet-50 · COCO 5-shot | AP .5:.95: +2.6 | Helped |
| Add DAP pre-training rather than classification pre-training | Faster R-CNN FPN ResNet-50 · COCO 5-shot | AP .5:.95: +4.7 | Helped |
| Add DAP pre-training rather than classification pre-training | Faster R-CNN FPN ResNet-50 · VOC 5-shot | AP .5: +13.4 (IN-1M); +19.5 (IN-14M) | Helped |
| Add full-network feature adaptation during DAP detector pre-training instead of freezing ResNet-50 (ablation) | Faster R-CNN FPN ResNet-50 · COCO 2017 val | AP .5:.95: 36.89 -> 37.25 | Helped |
| Use multi-threshold CAM box generation instead of any single tested threshold (ablation) | DAP Faster R-CNN FPN ResNet-50 · COCO 2017 val | AP .5:.95: 36.89, 36.82, 36.90, 36.97 -> 37.25 | Helped |
| Vary the pseudo-box NMS IoU threshold from 0.8 to the tested alternatives (ablation) | DAP Faster R-CNN FPN ResNet-50 · COCO 2017 val | AP .5:.95: 37.25 (0.8); 37.26 (0.9); 37.12 (0.5); 36.98 (0.6); 37.13 (0.7); 37.02 (1.0) | No clear effect |
Closest papers by meaning
- Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection (CVPR 2020 · similarity 0.88)
- Unsupervised Object Detection Pretraining with Joint Object Priors Generation and Detector Learning (NeurIPS 2022 · similarity 0.88)
- Weakly Supervised Object Localization With Progressive Domain Adaptation (CVPR 2016 · similarity 0.87)
- W2F: A Weakly-Supervised to Fully-Supervised Framework for Object Detection (CVPR 2018 · similarity 0.87)
- ELDET: Early-Learning Distillation with Noisy Labels for Object Detection (NeurIPS 2025 · similarity 0.86)
- Training Object Class Detectors With Click Supervision (CVPR 2017 · 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.