ObjectBox: From Centers to Boxes for Anchor-Free Object Detection

ECCV 2022 (Oral Presentation)

Dependencies

PyTorch

This code is tested under Ubuntu 18.04, CUDA 11.2, with one NVIDIA Titan RTX GPU.
Python 3.8.8 version is used for development.

Preparation

Set the 'PATH' in '/data/coco.yaml' and '/data/VOC.yaml'
Set the 'project' flag in flag_sets.py

Training

Set 'task' flag in flag_sets.py as: 'train'

For MS-COCO 2017 experiments, set:
exp = 'coco'
in flag_sets.py

For PASCAL VOC 2012 experiments, set:
exp = 'pascal'
in flag_sets.py

Run train.py

Test

Set 'task' flag in flag_sets.py as: 'test'

Run val.py

Pretrained Checkpoints

Trained model on COCO can be found here.

AP_0.5:0.95	AP_0.5	AP_0.75	AP_S	AP_M	AP_L	AR₁	AR₁₀	AR₁₀₀	AR_S	AR_M	AR_L
46.8	66.4	50.4	28.7	51.8	61.1	36.9	58.8	63.0	44.5	68.0	78.6

Acknowledgements

This project is supported by Geotab Inc., the City of Kingston, and the Natural Sciences and Engineering Research Council of Canada (NSERC)

Citation

Please cite our papers if you use code from this repository:

@article{zand2022objectbox,
  title={ObjectBox: From Centers to Boxes for Anchor-Free Object Detection},
  author={Zand, Mohsen and Etemad, Ali and Greenspan, Michael},
  booktitle={European conference on computer vision},
  pages={1--23},
  year={2022},
  organization={Springer}
}

@article{zand2021oriented,
  title={Oriented bounding boxes for small and freely rotated objects},
  author={Zand, Mohsen and Etemad, Ali and Greenspan, Michael},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={60},
  pages={1--15},
  year={2021},
  publisher={IEEE}
}

Reference

Many utility codes are borrowed from YOLO.

ChangGiMoon/ObjectBox