This repository accompanies the manuscript entitled "Automated Location Invariant Animal Detection In Camera Trap Images Using Publicly Available Data Sources" available on bioArxivs: It is currently undergoing peer review, and this link will be updated once the peer review process is complete.
This repository contains datasets and bounding box annotations for the following species:
- Suidae (pig, boar, warthog)
- Striped hyena (Hyaena hyaena)
- Rhinoceros
Altogether nine datasets are available for download. For more information on the characteristics of the datasets, please refer to the publication.
The code on this repository allows for training RetinaNet for Location Invariant animal detection.
Jupyter notebooks:
- SSIM Image Sorter: Use to sort datasets based on their similarity
- Auto-Annotator: use to automatically annotate your own custom datasets.
- Training RetinaNet: use to train RetinaNet
- Optimization via Infusion: use to infuse FiN training with camera trap data
Datasets, annotations and Python scripts for training of deep learning models for camera trap image processing