Cows2021

This repository contains the source code that accompanies our paper "Towards Self-Supervision for Video Identification of Individual Holstein-Friesian Cattle: The Cows2021 Dataset" at: https://arxiv.org/abs/2105.01938. At its core, the code in this repository is adapted and extended (with permission) from William Andrew et al's work on "Visual Identification of Individual Holstein Friesian Cattle via Deep Metric Learning" published in Computers and Electronics in Agriculture 2021 -- paper, source code

The trained weights from the for cow detection and identification are included in this repository at weights/.

1. Clone this repository.
2. Modify the prefix in conda_detection.yaml with your own path and name. Create your environment: conda env create -f conda_detection.yml
3. Instead of installing pycocotools, use the pycocotools in this repository(Detection/keras_retinanet/bin/cocoapi).

To test a trained model, copy the trained model weight for detection resnet50_trained_144.h5 to Detection/test/trained_model. Run rotate_test.py. An example would be python rotate_test.py --confidence_score_th=0.3 --nms_threshold=0.28 --save_path='xxx' --set_dir='xxx' --model_path='trained_model/resnet50_trained_144.h5' --print_id=0 --show_time=0 --pre=1 --map=1 --withground=1

Go to Detection/keras_retinanet/b_rotated, run train_rotated.py. An example would be python train_rotated.py --epochs=1 --steps=2 --evaluation=1 --snapshot_path='snapshots' --freeze_backbone=0 --random_transform=1 --dataset_path='xxx'

Download training data from Sub-levels/Identification/Train. Alternatively, you can generate the trainning data from raw videos from Sub-levels/Identification/Videos. Run the code in make_data one by one. You can found training images in a folder called Crop_split. When running the codes, you will need the trained model weight of detection, which can be found in Sub-levels/3Weights/trained_model/resnet50_trained_144.h5.

To test a trained model by inferring embeddings and using GMM to obtain the accuracy, run the code in Test one by one.

To train the model, use python train.py -h to get help with setting command line arguments. A minimal example would be python train.pyxxx

Consider citing ours and William's works in your own research if this repository has been useful:

@article{gao2021towards,
  title={Towards Self-Supervision for Video Identification of Individual Holstein-Friesian Cattle: The Cows2021 Dataset},
  author={Gao, Jing and Burghardt, Tilo and Andrew, William and Dowsey, Andrew W and Campbell, Neill W},
  journal={arXiv preprint arXiv:2105.01938},
  year={2021}
}

@article{andrew2020visual,
  title={Visual Identification of Individual Holstein Friesian Cattle via Deep Metric Learning},
  author={Andrew, William and Gao, Jing and Campbell, Neill and Dowsey, Andrew W and Burghardt, Tilo},
  journal={arXiv preprint arXiv:2006.09205},
  year={2020}
}