You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Apr 21, 2025. It is now read-only.
Tests revealed that one of the test sequences (Netflix's DinnerScene) causes perceptual objective metrics (VMAF/SSIM/MS-SSIM) based BD-rates to max out for HEVC on Gen 9, as x265-veryslow reference (tuned for PSNR) has a significant negative shift for this outlier sequence with film-grain noise. The VMAF tool based BD-rate calculator is maxing out the BD-rate at +-1M% which causes the BD-rate test set average to be significantly skewed. This occurrence needs to be handled better. A solution could be to assign maxed out BD-rate values in the results an "N/A" label and skip it in calculating BD-rate average.
Tests revealed that one of the test sequences (Netflix's DinnerScene) causes perceptual objective metrics (VMAF/SSIM/MS-SSIM) based BD-rates to max out for HEVC on Gen 9, as x265-veryslow reference (tuned for PSNR) has a significant negative shift for this outlier sequence with film-grain noise. The VMAF tool based BD-rate calculator is maxing out the BD-rate at +-1M% which causes the BD-rate test set average to be significantly skewed. This occurrence needs to be handled better. A solution could be to assign maxed out BD-rate values in the results an "N/A" label and skip it in calculating BD-rate average.