AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Self Supervised Learning, Comparative Learning
We started from stereo set up with two cameras to just a singular camera. And actually we've worked on the hardest problem first, which is monocular camera. The reason we've done that is because it tha common denominator behind almost any platform. We teach them by reconstruction, using geometry as an equation in the middle to get the prediction. And we learn using ad back prop and the way we do it. So what we found is, why does self surprise learning and particulaly contrastive learning work so well?