I need to corroborate the convention used by Fastai2 to describe a tensor image. First, it is my understanding that the coordinate space for a tensor image is -1,-1 to +1,+1. Now, back in Fastai1, the coordinate convention to describe a point in this space was yxyx. Is it the same in Fastai2? or has it changed to xyxy? Thanks for your help
Its xyxy normalized to a -1,1 scale for tensor points. See my keypoints notebook here: https://github.com/muellerzr/Practical-Deep-Learning-for-Coders-2.0/blob/master/Computer%20Vision/06_Keypoint_Regression.ipynb
(Edited @joseadolfo, apologies)
It’s done so we can apply augmentation easily.