Hence, the VMAF scores are usually linked closer to the actual perceived quality of a video.
While PSNR is based on the mean squared error, VMAF is a machine learning-based model that was trained on actual MOS scores. This can falsify the results as it leads to a frame shift and an unequal number of frames between source- and encoded video. Some encoders include an additional black frame at the beginning of the video. Usually, the overall score is based on the arithmetic mean, alternatives like the harmonic mean are also valid and possible. This means that videos are compared frame by frame and compute an overall score. Both video quality metrics work on a frame basis. In order to compute a PSNR or a VMAF score, the source video and the encoded video need to have the same resolution. Although I will not go into the details of both metrics, I want to highlight some of the key facts and common pitfalls: Based on the source video, and the encoded video a video quality score is derived. Video quality metrics such as VMAF and PSNR promise to do exactly that. Ideally, this process is completely automated and does not require any human interaction. Hence, we need a more sophisticated approach to determine the quality of a video. Even Netflix and YouTube can not afford to put thousands of people in front of a TV and let them rate the quality of their encodes. However, it requires us to watch all of our encodes multiple times, which obviously only scales to a certain degree. This simple scoring system is often referred to as the Mean Opinion Score (MOS). For instance, we can rate the quality of our encodes on a scale from 1 to 5, with 1 being bad and 5 being very good. The most straightforward way to determine the quality of a video is by watching it and assigning a subjective quality score. As an example, this is a mandatory step for per-title encoding, which I explained in one of my previous blog posts. Let us assume that we’ve created multiple encodes from our high quality source video and would like to check if the quality of our encodes is sufficient enough to deliver the video to the end user. In this blog post, I will explain how to calculate the Video Multi-Method Assessment Fusion (VMAF) and Peak signal-to-noise-ratio (PSNR) scores in a single FFmpeg command, and produce the results in JSON format or a text file. Depending on the concrete use case, however, it can be challenging to assemble the right command. FFmpeg is a great tool for video processing, it basically allows us to manipulate videos any way we like.
0 kommentar(er)
