Drone Bird Detector

Trained model for detecting and identifying birds in drone imagery, accompanying:

Wilson, J. P., T. Amano, T. Bregnballe, A. Corregidor-Castro, R. Francis, D. Gallego-García, et al. (2026). Big Bird: A global dataset of birds in drone imagery annotated to species level. Remote Sensing in Ecology and Conservation. https://doi.org/10.1002/rse2.70059

Trained on:

Insert Dataset DOI.


Model Overview

This model performs object detection and fine‑grained classification of birds in aerial drone images. It predicts hierarchical labels including:

  • Unique label ID
  • Taxonomic information: class → order → family → genus → species
  • Traits: age category (chick, juvenile, adult), sex (male, female, monomorphic)
  • Combined into a full label like:
1;aves;procellariiformes;diomedeidae;thalassarche;melanophris;adult;monomorphic;black-browed albatross - adult - monomorphic

Training Images: high‑resolution drone RGB imagery. Model Type: Faster RCNN with resnet-101 backbone.


Intended Use

Primary use cases

  • Automated processing of large drone image collections for ecological surveys
  • Generating species counts and spatial patterns from aerial surveys
  • Supporting conservation monitoring and research workflows

Not intended for

  • Real‑time inference on live video streams
  • Predicting species not present in the training dataset
  • Making decisions without ecological validation

Limitations & Ethical Considerations

This model reflects biases in the training data (e.g., species representation, habitat contexts). Users should:

  • Validate outputs with domain experts
  • Be cautious of false positives in complex scenes
  • Understand that model performance may vary with altitude, sensor quality, and lighting

Evaluation

Model performance metrics (precision/recall, mAP, per‑class accuracy, etc.) were reported in Wilson et al. (2026). Users are encouraged to evaluate the model on their own images and share results.


How to Download

You can use the huggingface_hub Python client to download model weights and labels:

from huggingface_hub import hf_hub_download

# Download model weights
model_path = hf_hub_download(
    repo_id="JoshuaWilson/drone-bird-detector",
    filename="model.pth"
)

# Download label definitions
labels_path = hf_hub_download(
    repo_id="JoshuaWilson/drone-bird-detector",
    filename="labels.txt"
)

Citation

Please cite the original paper when using this model:

@article{Wilson2026,
  title={Big Bird: A global dataset of birds in drone imagery annotated to species level},
  author={Joshua P. Wilson and Tatsuya Amano and Thomas Bregnballe and Alejandro Corregidor-Castro and Roxane Francis and {Diego Gallego-García} and Jarrod C. Hodgson and Landon R. Jones and {César R. Luque-Fernández} and Dominik Marchowski and John McEvoy and Ann E. McKellar and W. Chris Oosthuizen and Christian Pfeifer and Martin Renner and {José Hernán Sarasola} and {Mateo Sokač} and Roberto Valle and Adam Zbyryt and Richard A. Fuller},
  journal={Remote Sensing in Ecology and Conservation},
  year={2026},
  doi={10.1002/rse2.70059}
}

License

This model is released under the MIT License.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support