in train.py:
comment says # Downscale (bicubic interpolation)
but according to skimage documentation
skimage.transform.resize is using order=1 by default which is bilinear(see warp )
i tried cv2.resize instead of scipy.interp
cv2.resize reaches a psnr performance of 35+dB which is normal.
though scipy.interp can only reach 33+dB (current implementation)
i think there is a problem here.
in train.py: comment says # Downscale (bicubic interpolation) but according to skimage documentation skimage.transform.resize is using order=1 by default which is bilinear(see warp )