COw LOcalization (COLO) Dataset

The COw LOcalization (COLO) dataset is designed to localize cows in various indoor environments using different lighting conditions and view angles. This dataset offers 1,254 images and 11,818 cow instances, serving as a benchmark for the precision livestock farming community.

Dataset Configurations

Configuration	Training Split	Testing Split
0_all	Top-View + Side-View	Top-View + Side-View
1_top	Top-View	Top-View
2_side	Side-View	Side-View
3_external	External	External
a1_t2s	Top-View	Side-View
a2_s2t	Side-View	Top-View
b_light	Daylight	Indoor + NIR
c_external	Top-View + Side-View	External

Download the Dataset

To download the dataset, you need to have the required Python dependencies installed. You can install them using either of the following commands:

python -m pip install pyniche

or

pip install pyniche

Once the dependencies are installed, use the Python console to provide the download destination folder in the root parameter and specify the export data format in the format parameter:

from pyniche.data.download import COLO

# Example: Download COLO in the YOLO format
COLO(
    root="download/yolo",  # Destination folder
    format="yolo",  # Data format
)

# Example: Download COLO in the COCO format
COLO(
    root="download/coco",  # Destination folder
    format="coco",  # Data format
)

Citation

The page of the arXiv article

@misc{das2024model,
    title={A Model Generalization Study in Localizing Indoor Cows with COw LOcalization (COLO) dataset},
    author={Mautushi Das and Gonzalo Ferreira and C. P. James Chen},
    year={2024},
    eprint={2407.20372},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

or

Das, M., Ferreira, G., & Chen, C. P. J. (2024). A Model Generalization Study in Localizing Indoor Cows with COw LOcalization (COLO) dataset. arXiv preprint arXiv:2407.20372

---