Global Wheat Detection

Code for Global Wheat Detection competition, hosted on kaggle.com.

Notes

Configuration

Train script uses Hydra framework from Facebook Research.

To change various settings you can either edit .yaml files in the config folder or pass corresponding params to the command line. The second option is useful for quick testing. For example:

python src/train.py train.epoch_length=20 logging.iter_freq=10

For more information please visit Hydra docs.

Usage

Multi-GPU training

Launch distributed training on GPUs:

python -m torch.distributed.launch --nproc_per_node=2 --use_env src/train.py

It's important to run torch.distributed.launch with --use_env, otherwise hydra will yell at you for passing unrecognized arguments.

Requirements

OS: Ubuntu 18.04.4 LTS (5.0.0-37-generic)
CUDA 10.1.243, driver 435.21
Conda 4.8.3
Python 3.7.7
PyTorch 1.4.0

Citations

@inproceedings{jackson2019style,
  title={Style Augmentation: Data Augmentation via Style Randomization},
  author={Jackson, Philip T and Atapour-Abarghouei, Amir and Bonner, Stephen and Breckon, Toby P and Obara, Boguslaw},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
  pages={83--92},
  year={2019}
}

@Inproceedings{zheng2020distance,
  author    = {Zhaohui Zheng, Ping Wang, Wei Liu, Jinze Li, Rongguang Ye, Dongwei Ren},
  title     = {Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression},
  booktitle = {The AAAI Conference on Artificial Intelligence (AAAI)},
   year      = {2020},
}

@misc{solovyev2019weighted,
    title={Weighted Boxes Fusion: ensembling boxes for object detection models},
    author={Roman Solovyev and Weimin Wang},
    year={2019},
    eprint={1910.13302},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}