An Intriguing Failing of Convolutional Neural Networks and the CoordConv Solution

Uber labs AI labs recently released a blog post describing how convolutional networks fail in very simple tasks mapping a single pixel into a 2d grid of coordinates. They even took it further to find out if the same results hold for Conv GANs as well. Wanna know what happened?

I suggest you read the whole article for further details!

Leave a Reply

Your email address will not be published. Required fields are marked *