The Machine Learning Tokyo Hackathon: Face Recognition was held on April 21 at Tokyo Chapter, Roppongi. It was limited to 15 seats and Machine Learning and Deep Learning engineers joined to build face detection, face recognition and other related systems.
We started with exploring OpenCV and Haarcascades, which is usually the first approach within the field. Face detection worked relatively well with both, static images as well as real time with a webcam. However, we found sometimes false positives in the corner of the lips or the nose, that was incorrectly recognized as eyes.
One of the participants built a face recognition system that was recognizing only his face, even among 5 or more people on screen.
After playing with around OpenCV, one participant explored CoreML, a platform that allows you to integrate trained machine learning models into an iOS app. He used the Inception model, a object detection model trained to classify 1000 different objects, and deployed it to a simple app prototype.
You can find the initial OpenCV src code we used on out official Github (that we’re just building up).
Our next event will be held at the end of May. We’re organizing a hands-on workshop with two instructors on Generative Adversarial Networks (GANs). The working session is limited to 15 people and will be announced on Meetup.