libaom is an AV1 software video encoder and decoder that functions both as the reference implementation, and is used in production to encode content such as AVIF images.
I ported the "still image" tune from SVT-AV1-PSY to libaom, resulting in improved visual quality and better scores on modern image quality metrics, including SSIMULACRA 2, DSSIM, FLIP, and Butteraugli. Users can generally expect a 10-20% reduction in file size for photographs. You can find the commit history and details of my changes here.
SVT-AV1-PSY is an AV1 software video encoder born from the SVT-AV1 project. Our encoder is designed for optimal perceptual fidelity; we have optimized heavily around pleasing the human visual system, and we aim to provide the best perceptual fidelity available in any AV1 video encoder across a wide quality spectrum.
As one of the main contributors of this project, I added features that improve quality of life, and encode quality and consistency. Most notably, I co-developed a new "still image" tune with Gianni Rosato that improves visual quality for AVIF images, as well as removing important resolution limitations. Visual comparisons can be found here.
SVT-AV1 is an open-source software AV1 encoder that is architected to yield excellent quality-speed-latency tradeoffs on CPU platforms for a wide range of video coding applications.
I worked on upstreaming my Variance Boost feature to mainline SVT-AV1, including research and design, implementation, writing documentation, and bug-fixing.
One of my goals is to incrementally incorporate SVT-AV1-PSY features back into the SVT-AV1 project, so more people can enjoy the benefits of our improvements.
At Azure DevOps, I worked with various teams and products. The most notable features I contributed were:
At GitHub, I worked for the GitHub Actions team on these things (among others):