Traditional video compression methods are designed to minimize bitrate and distortion. The goal is to achieve the lowest possible bitrate with the least possible perceived distortion. Typically, the distortion to be minimized is measured using the Peak Signal-to-Noise Ratio (PSNR). However, it is also well known that PSNR does not correlate particularly well with human vision.
In addition to optimizing PSNR, modern video compression methods also offer other metrics for minimizing noise (known as “tuning modes”). Scalable Video Technology for AV1 (SVT-AV1), for example, recently added Video Multi-Method Assessment Fusion (VMAF) as a tuning mode.
The goal of this work is to provide a visual comparison of the available tuning modes in SVT-AV1 using various independent video quality metrics, as well as an analysis of encoding times and the coding tools used across the different tuning modes.
Prerequisites
- Solid background in image and video compression
- Experience in Python
- Interest in video compression and video quality