Comment by cs702

The OP seemingly forgot to mention the fact that using CoordConv with GANs results in more realistic generation of images, with smooth geometric transformations (including translation and deformations) of objects. Examples:

* https://eng.uber.com/wp-content/uploads/2018/07/image5.gif

* https://eng.uber.com/wp-content/uploads/2018/07/image11.gif

* https://eng.uber.com/wp-content/uploads/2018/07/image12.gif

These and other examples suggest CoordConv can significantly improve the quality of the representations learned by existing architectures.

That doesn't seem so "trivial."