Researched lightweight deep learning approaches for video restoration in the AV2 codec, generating multiple models capable of increasing quality (PSNR gain) by up to 2.5%.
Developed a method to generate quantized weights for linear combinations of output channels, called encoder-guided restoration, leading to improvements in codec performance of up to 2%.
Investigated distillation as a method of achieving highly efficient and effective models for video restoration.
Evolution of codec, Ali266 unlocks new growth in the field of ultra-high-definition video
Codec • Source: Video Cloud Technology • February 1, 2023 at 5:04 pm • Technical Articles