Australian Camera Trap Data (ACTD)

The dataset comprises images from camera traps deployed across various sites in Victoria, Australia, regions to monitor biodiversity and conservation. The ecological experts manually sort and review the camera trap data based on the species class. The reviewed camera trap data is processed through the mega detector model to collect the bounding box coordinates of the species. The proposed dataset consists of three variants to address data imbalance issues in species classification by grouping species into higher-level categories (e.g., birds and small animals) called terrestrial grouped species data, region-specific species data and feral animal data. Each dataset has cropped animal images, YOLO annotated files and COCO formatted JSON files to train efficient deep learning models. All scripts used for data processing, annotation and validation are publicly available in the GitHub repository: GitHub - sameeruddin/ACTD_scripts.