Intel's video processing library “libvpl'' has released a new version ahead of the holiday season. The only major change in libvpl 2.16 is the addition of an experimental API for AI-assisted video encoding functionality.
The Intel libvpl library acts as a dispatcher to interface with various implementations of Intel Media SDK or Intel oneVPL or Intel VPL GPU Runtime components. The only changes listed in tonight's libvpl 2.16 release are:
“Support for Intel® VPL API 2.16 (includes new AI-assisted encoder API and documentation updates)”
Of course, the AI-assisted encoder API piqued my interest. I was curious about the new API, so I looked into it and found a new API. mfxExtAIEncCtrl Interface added as an experimental feature for controlling AI-assisted encoder solutions. The AI-assisted encoder solutions added to the specification as part of this are a saliency encoder solution and an ML-based adaptive target usage solution.
libvpl is just a dispatch library, so there are no further details. We have not yet seen any commits related to the actual implementation from the Intel VPL GPU Runtime, except that these AI-assisted features are reserved exclusively for some binary-only components. (the only mfxExtAIEncCtrl References in the published vpl-gpu-rt open source code were header structures, some dump/trace functions, and assert code. )
If you want to check out the Intel libvpl 2.16 release, you can find it on GitHub.
