CoordConv: An intriguing failing of convolutional neural networks and the CoordConv solution

August 2019

tl;dr: Predicting coordinate transformation (predicting x and y directly from image and vice versa) with Conv Nets are hard. Adding a mesh grid to input image helps this task significantly.

Overall impression

The paper results are very convincing, and the technique is super efficient. Essentially it only concats two channel meshgrid to the original input.

RoI10D cited this paper. This work also inspired cam conv.

Key ideas

Technical details