File size: 3,398 Bytes
8b4bf5b 9f95797 8b4bf5b 9f95797 8b4bf5b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 | ---
license: mit
tags:
- bird-detection
- ecology
- remote-sensing
- computer-vision
datasets:
- DOI
---
# 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:
```python
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:
```bibtex
@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**.
|