Google AI Depth Prediction with Single Camera

I also recently learned that Google AI recently developed a neural network to generate predictions regarding depth: http://ai.googleblog.com/2019/05/moving-camera-moving-people-deep.html. Interestingly, they trained the neural network on the hundreds of videos posted on YouTube pertaining to the Mannequin Challenge. The reason was that in the mannequin challenge, objects are stationary, enabling depth measurement using deterministic algorithms. They then train the neural network to match these values for a single frame. Another slightly more complex version of the model involved passing nearby frames before and after to the neural network. I think that's freakin' cool.

Comments

Popular posts from this blog

My Summer

Thoughts on LLMs and Modeling

A Realization