Hello, a more formal question, sorry.
I have a CNN and receive as output an image of the shape [1, 512, 512] and supervise it with a ground truth of the same shape. Furthermore, I calculate MAE, MSE, SSIM, … and also the MedAE (Median-Absolute-Error). Formula (N pixels in each tile):
The question is for the latter index: When I then have an array with all my median errors, do I take the median from this array for result presentation or the mean of all median errors?