Research map / Detection and segmentation / Weakly supervised detection and segmentation
DETReg: Unsupervised Pretraining With Region Priors for Object Detection
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 |
|---|---|---|---|
| pretrain the full detector using region-prior localization and object-embedding targets | DETR · MS COCO val2017, 150 epochs | AP: 40.5 -> 41.9 | Helped |
| pretrain with DETReg instead of SwAV backbone pretraining | DDETR · PASCAL VOC test07 | AP: 61.0 -> 63.5 | Helped |
| pretrain with DETReg instead of supervised backbone pretraining | DDETR · Airbus Ship Detection test | AP: 79.8 -> 81.0 | Helped |
| pretrain with DETReg rather than SwAV | DDETR · COCO 1% labeled data | AP: 11.79 $\pm$ 0.3 -> 14.58 $\pm$ 0.3 | Helped |
| pretrain with DETReg rather than SwAV | DDETR · COCO 2% labeled data | AP: 16.02 $\pm$ 0.4 -> 18.69 $\pm$ 0.2 | Helped |
| pretrain with DETReg rather than SwAV | DDETR · COCO 5% labeled data | AP: 22.81 $\pm$ 0.3 -> 24.80 $\pm$ 0.2 | Helped |
| pretrain with DETReg rather than SwAV | DDETR · COCO 10% labeled data | AP: 27.79 $\pm$ 0.2 -> 29.12 $\pm$ 0.2 | Helped |
| use DETReg pretraining instead of supervised pretraining | DDETR · COCO few-shot novel classes, 30 shots | Novel AP: 20.4 -> 22.6 | Helped |
| use DETReg pretraining rather than supervised pretraining | DDETR · COCO few-shot without base-class training, 10 shots | Novel AP: 6.7 -> 9.1; 10.0 vs 9.1 | Mixed |
| shuffle region proposal box locations across images (ablation) | DETReg · MS COCO finetuning, IN100 pretraining | AP: 45.1 -> 32.0 | Hurt |
| add the object embedding loss with $\lambda_e=1$ (ablation) | DETReg · MS COCO finetuning, IN100 pretraining | AP: 43.3 -> 45.1 | Helped |
| freeze the backbone during pretraining (ablation) | DETReg · MS COCO finetuning, IN100 pretraining | AP: 45.1 -> 45.4 | No clear effect |
Closest papers by meaning
- UP-DETR: Unsupervised Pre-Training for Object Detection With Transformers (CVPR 2021 · similarity 0.89)
- MS-DETR: Efficient DETR Training with Mixed Supervision (CVPR 2024 · similarity 0.88)
- UN-DETR: Promoting Objectness Learning via Joint Supervision for Unknown Object Detection (AAAI 2025 · similarity 0.87)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.86)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.86)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 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.