Brooks Falls

This page provides an overview of hypraptive’s open source BearID Project, started in January of 2017. The goal of the project is to identify individual brown bears from photos and videos; face recognition technology for bears! The project resulted from the collision of two interests: deep learning and the Brooks Falls bearcams.

We started out by focusing on the Bears of Brooks River. We first needed to understand how the experts identify individual bears. Next, we reviewed some existing deep learning and computer vision projects aimed at identifying animals in the wild. We decided on an initial approach based on FaceNet, but for bears. After trying different software packages and methods (YOLO, dlib, HOG and the dog hipsterizer), we came out with BearID 1.0.

It was a good start, but we still needed more data to improve accuracy.

Brown Bear Research Network

We are working on the BearID project in collaboration with The Brown Bear Research Network (BBRN). We have the common goal of developing a new technology which can be used to assess and monitor populations, providing wildlife researchers with a new methodology to survey bears in the wild. Hypraptive provides the computer science and deep learning experience and the BBRN provides brown bear expertise and connections to the bear viewing community. As well as being a vital research tool, the ability to automatically and confidently recognize individual bears has huge potential for public engagement in bear conservation.

Read more about how this collaboration came to be in our post: Kindred Spirits.

Our Datasets

You can get an overview of our bear face datasets in the following posts:

We need YOUR HELP!

Bear Face Chips

Deep-learning methods require thousands of images to achieve the high accuracy of recognition required. We are looking for brown bear photographers, researchers, managers and enthusiasts to donate images and video footage to the project to enable us to improve the accuracy of our system. Supplied images and videos will be used for the sole purpose of training and testing the system.

Photos/videos can be of wild or captive brown bears, from anywhere in the world, provided they can be uniquely identified. Any resolution is ok as long as both eyes of the bear are clearly visible. A unique bear number/name should be in each file name or images of each bear can be grouped in folders (example: brooks0032). Please submit your images/footage or questions to photos@bearresearch.org (include a dropbox link or similar if required) or we can mail you a flash/hard drive.

Keep track of our progress via the hypraptive blog. Follow the Brown Bear Research Network on twitter (@bear_network) and Facebook (@bearresearch).

Thank you in advance to the bear community for your help!

Call for Help