Optimization of a video transcoder is performed, e.g., by fine-tuning their parameters, based on evaluation of the performance of a transcoder over a small, fixed video dataset. The use of a small, fixed video dataset enables reproducibility, fast evaluation, and regression testing. However, transcoders that are fine-tuned based on a small, fixed dataset can often deliver suboptimal transcoding performance when utilized to transcode videos from a much larger dataset, e.g., videos served by a video hosting and sharing service. This is because a small, fixed set of videos is not sufficiently representative of the total corpus of videos hosted by a video sharing service and does not cover the scale and diversity of such videos. This disclosure describes the use of importance sampling in the evaluation of video transcoders using a small dataset of videos. The techniques can deliver a high-performance transcoder even when the transcoder is optimized using a small dataset that is insufficiently representative of a large-scale video corpus.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.