Wild Animal Facing Extinction

Wildlife conservation is an essential matter, particularly in Africa, where hunting and poaching constitute an important danger to rhinos and elephants. Despite the efforts of governments and NGOs, the scarcity of annotated wild animal datasets has hindered the development of computer algorithms for monitoring and protecting these animals in natural reserves and national parks. While advances in technology have enabled institutions to upload images for monitoring purposes, manual annotation remains time-consuming and inefficient. To address this issue, the paper proposes a semi-automatic annotation framework for wild animal dataset construction, focusing on six classes (rhino, elephant, lion, giraffe, cheetah, and zebra). The framework involves training a model on a small manual dataset and using this to semi-automatically label the remaining large-scale dataset. Additionally, the authors manually annotated rhino images with skeletons to be expropriated in computer vision algorithms for pose estimation tasks. This is the first wild animal dataset constructed with boundary box labels based on semi-automatic annotation and rhino skeletons manually annotated. The proposed framework has the potential to expedite research in wildlife protection, serve as a guide for semi-automatic annotation dataset construction, and possess a significant impact on the conservation community. Our dataset is publicly available here:
Keywords: Wild animal dataset, semi-automatic annotation, deep learning algorithms, boundary box labels, rhino skeleton, time efficient

Wild Animals Boundary Box Labels
Class Number of Images Train Valid Test
Rhino 1389 6107 763 764
Cheetah 1150
Elephant 1288
Lion 1277
Zebra 1321
Giraffe 1209
Total 7634

Image = .jpg & Annotation = .txt

Rhino Skeleton Labels
Class Number of Images Train Valid Test
Rhino Skeleton 1410 1048 262 100
Total 1410

Image = .jpg & Annotation = .json

NB: The Rhino Skeleton dataset was annotated manually on this site V7:V7.https://www.v7labs.com and can be separately cited as below:
Citation - Video Surveillance based Wild Animal Poaching Mitigation (2023).Rhino Skeleton.
BAKANA SIBUSISO REUBEN, V7 Open Datasets. https://darwin.v7labs.com/video-surveillance-based-wild-animal-poaching-mitigation/rhino-skeleton
Also is it publicly available at: https://darwin.v7labs.com/video-surveillance-based-wild-animal-poaching-mitigation/rhino-skeleton