Unable to reproduce experimental results in the paper. #27
Description
I downloaded the UVG dataset with 1080p video clips and tested it with a pre-trained 2048.model, but was unable to reproduce the experimental results in the paper.
Here is the relevant data:
Ipp=[1.2929020996093752, 0.94005859375, 0.7543700358072918, 0.6924034016927084, 0.6758680826822915, 0.6770526529947917, 0.8640651041666668]
test result:[ average bpp : 0.174915, average psnr : 36.798520, average msssim: 0.948412]
Alternatively, we trained ourselves to produce results that are not far from this one.
I'd like to ask if there's a problem with my data processing or with the training methodology.