AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
How to Train Optical Flow Estimation Networks With Real Data
The cost volume that you mentioned is it sounds like roughly multi dimensional error metric. So this is one way of doing that. Of course, earlier AI based solutions again, deep learning solutions, but they will do they will take two frames and test through a feed forward network without any cost volume. But at the end they will predict the delta x delta y these motion directions for each pixel. We have very limited data, except synthetic data coming from, let's say, game engines. It's in a way doesn't require, you know, any label data to train optical flow solutions which is very innovative. That's what we showed at CBPR, but it does it takes this any video