no code implementations • 24 Mar 2020 • Liliang Ren, Zhuonan Hao
Thus the generalization ability of CNN will be limited since the coordinate information is crucial for a model to learn affine transformations which directly operate on the coordinate of each pixel.