“The final results from Great Elephant Census show 352,271 African savanna elephants in 18 countries, down 30% in seven years.”
- The Great Elephant Census
A critical challenge to wildlife preservation is the protection of highly endangered animals. The African elephant, the largest land mammal in the world, is a keystone species that plays an outsized role in maintaining the fragile ecosystem in which they live. Due to a booming ivory trade, African elephants in 18 countries have found their numbers decline from an estimated 3-5 million in the 1930s, to roughly 350K remaining today.
One of the key challenges in this conservation is simply counting them. In order to make a political case for saving elephants and to ensure that the policies we enact are working, we need to be able to track their numbers effectively. This is where computer vision can be a powerful tool to help with this plight.
WHAT WE DID
To prove the value of our synthetic data, we tested a state-of-the-art object detection algorithm called YOLO-v2 (You Only Look Once) on drone footage of real elephants roaming a grassland in Africa.
WHAT WE LEARNED
Microsoft COCO Real Data
We took a set of pre-trained weights for this algorithm, (already trained on a large dataset called MS (Microsoft) COCO which contains data across 80 object classes including elephants), and tested it on this video to obtain results.
The algorithm constantly misidentifies an object as a bird, sheep or cow, but rarely an elephant. It turns out that the MSCOCO dataset does not contain many images of elephants from an aerial perspective.
AI.Reverie Synthetic Data
We further reviewed the data to understand why this might be the case. We then created another scenario where we train YOLO from scratch, but included our AI.Reverie synthetic data which contains aerial perspectives of elephants.
Our synthetic data of elephant images at different perspectives augmented the training of the algorithm and greatly enhanced its ability to detect elephants.