Object Detection in adverse weather

📢 This project is based on the following GitHub: GitHub - WoojuLee24/OA-DG

OA-DG Method Summary

Object-Aware Domain Generalization
Type: Single Domain
Base Model : Faster R-CNN (2-stage Detector) with ResNet-101 backbone
Method : Image augmentation, Domain Generalization
Dataset : DWD, Cityscapes

• Cityscapes: A dataset that contains urban street scenes from 50 cities with detailed annotations.
• Diverse Weather Dataset: This dataset includes various weather conditions for robust testing and development of models, essential for applications in autonomous driving.
  Collected Data from BDD-100k(2020), FoggyCityscapes(2018) and Adverse Weather(2020).

Train

DWD dataset (Diverse Weather Dataset)

python tools/train.py configs configs/OA-DG/dwd/faster_rcnn_r101_dc5_1x_dwd.py --work-dir /home/intern/minkyoung/dataset/DWD/faster_rcnn_r101_dc5_1X_dwd_oadg --gpu-ids 3

Cityscapes dataset

Used classes -> ('person', 'car', 'truck', 'bus', 'motorcycle', 'bicycle')

python -u tools/train.py configs/OA-DG/cityscapes/faster_rcnn_r50_fpn_1x_cityscapes_oadg.py --work-dir /home/minkyoung/dataset/cityscapes/faster_rcnn_r50_fpn_1x_cityscapes_oadg/exp2

---

Evaluatioin

cityscapes dataset

python -u tools/analysis_tools/test_robustness.py configs/OA-DG/cityscapes/faster_rcnn_r50_fpn_1x_cityscapes_oadg.py /home/intern/minkyoung/dataset/cityscapes/faster_rcnn_r50_fpn_1x_cityscapes_oadg/epoch_2.pth --out /home/intern/minkyoung/dataset/cityscapes/faster_rcnn_r50_fpn_1x_cityscapes_oadg/test_robustness_result_2epoch.pkl --corruptions benchmark --eval bbox

---

Reference

GitHub - AmingWu/Single-DGOD
OA-DG Paper