Deep Encode – Towards Context Aware Encoding
(Paper available in the Proceedings of the 2020 BEIT Conference, under the title, "Machine Learning for Per-Title Encoding")
Video streaming content differs in terms of complexity and requires title-specific encoding settings to achieve a certain visual quality. Classic ”one-fits-all” encoding ladders ignore video-specific characteristics and apply the same encoding settings to all video files. In the worst-case scenario, this approach can lead to quality impairments, encoding artifacts or unnecessary large media files. A per-title encoding solution has the potential to significantly decrease the storage and delivery costs of video streams while improving the perceptual quality of the video. Traditional per-title encoding solutions typically require a large number of test encodes, resulting in high computation times and costs. In this paper, we illustrate a solution that implements the traditional per-title encoding approach and uses the resulting data for machine-learning-based improvements. By applying supervised, multivariate regression algorithms like Random Forest Regression, Multilayer Perceptron and Support Vector Regression we are able to predict mandatory video quality metrics (VMAF). That way, the test encodes are eliminated while preserving the benefits of per-title encoding.
This session will be rebroadcast on the BEIT Express channel on May 14, 2020 at 1:30 a.m. and 9:30 a.m. EDT (UTC -4).