Research map / Detection and segmentation / CNN features
Bootstrap Your Object Detector via Mixed Training
Mengde Xu, Zheng Zhang, Fangyun Wei, Yutong Lin, Yue Cao, Stephen Lin, Han Hu, Xiang Bai
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 |
|---|---|---|---|
| replace single-input training with MixTraining | Faster R-CNN ResNet-50 · COCO2017 val | mAP: 41.7 -> 44.0 | Improved |
| replace SiTraining with MixTraining on Swin-S backbone | Faster R-CNN Swin-S · COCO2017 val | mAP: 48.7 -> 50.3 | Improved |
| replace SiTraining with MixTraining | Cascade R-CNN Swin-S · COCO2017 val | mAP: 50.9 -> 52.8 | Improved |
| extend training from 360K to 720K iterations with MixTraining | Faster R-CNN ResNet-50 · COCO2017 val | mAP: 42.4 -> 44.0 | Improved |
| extend SiTraining (normal aug) from 360K to 720K iterations | Faster R-CNN ResNet-50 · COCO2017 val | mAP: 41.7 -> 36.1 | Hurt |
| use strong augmentation in SiTraining and extend to 720K | Faster R-CNN ResNet-50 · COCO2017 val | mAP: 40.1 -> 38.9 | Mixed |
| add mixed data augmentation (ablation) | SiTraining · COCO2017 val | mAP: 41.7 -> 42.5 | Improved |
| add mixed training targets (pseudo boxes) on top of mixed augmentation (ablation) | MixTraining · COCO2017 val | mAP: 42.5 -> 44.0 | Improved |
| add pseudo boxes for missing labels (ablation) | MixTraining · COCO2017 val | mAP: 42.5 -> 43.7 | Improved |
| add pseudo boxes for localization noise only (ablation) | MixTraining · COCO2017 val | mAP: 42.5 -> 42.9 | Improved |
| use hybrid of both pseudo box strategies instead of missing label only (ablation) | MixTraining · COCO2017 val | mAP: 43.7 -> 44.0 | Improved |
| apply mixed augmentation strategy without weighted loss (ablation) | SiTraining · COCO2017 val | mAP: 41.7 -> 42.1 | Improved |
Closest papers by meaning
- Many-shot from Low-shot: Learning to Annotate using Mixed Supervision for Object Detection (ECCV 2020 · similarity 0.87)
- C-Mixup: Improving Generalization in Regression (NeurIPS 2022 · similarity 0.87)
- RecursiveMix: Mixed Learning with History (NeurIPS 2022 · similarity 0.87)
- Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond (AAAI 2024 · similarity 0.86)
- InstaGen: Enhancing Object Detection by Training on Synthetic Dataset (CVPR 2024 · similarity 0.85)
- MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection (ICLR 2024 · similarity 0.85)
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.
