I noticed that you're first slicing the Nifti volume and saving them as png images in [0, 255] range and training the png images by normalizing them. It looks like you're calculating MSE, MAE, PSNR, etc between two png/jpg images. Calculating MSE this way gives a different value than calculating MSE between two Niftii volumes since they have different ranges. I wonder what's the convention for calculating these metrics. Do people in the literature report the results of MSE, MAE, etc in terms of png/jpeg image or in terms of Nifti volume?
Wonderful work! :)
I noticed that you're first slicing the Nifti volume and saving them as png images in [0, 255] range and training the png images by normalizing them. It looks like you're calculating MSE, MAE, PSNR, etc between two png/jpg images. Calculating MSE this way gives a different value than calculating MSE between two Niftii volumes since they have different ranges. I wonder what's the convention for calculating these metrics. Do people in the literature report the results of MSE, MAE, etc in terms of png/jpeg image or in terms of Nifti volume?