AVS Deep Learning Video Coding Loop Filter Challenge Announcement

The "Deep Learning Based Video Coding Loop Filter Challenge" initiated by the Digital Audio‑Video Coding Technology Standard Working Group (AVS Working Group) and supported by iQIYI is officially launched.

The AVS Working Group was approved by the Ministry of Information Industry in June 2002. Its mission is to develop common technical standards for digital audio‑video compression, decompression, processing, and representation, meeting the needs of China’s information industry. The standards support high‑resolution digital broadcasting, high‑density laser digital storage, wireless broadband multimedia communications, and internet streaming. Since its inception, the group has produced two generations of AVS standards and is now collaborating with international video coding bodies on a new generation. At the June 2018 Hefei meeting, the baseline platform and technical direction were set, with the first phase of the new standard expected to be completed by the end of 2019, targeting more than 25% improvement in coding performance.

With the rapid rise of AI, deep‑learning‑based image and video coding algorithms have attracted increasing attention in recent years. Notably, a learning‑based image coding competition launched at a CVPR Workshop drew top research teams worldwide, demonstrating that the latest learning methods can achieve the same subjective quality with significantly lower bitrates. To advance the new video coding standard, the AVS video working group encourages research on novel coding technologies, especially learning‑based approaches. As a first step, the group has launched the "Deep Learning Based Video Coding Loop Filter Challenge". Loop filter technology is a critical component in video coding, essential for improving current frame quality and enhancing the prediction accuracy of subsequent frames. Experiments have shown that convolutional‑neural‑network‑based loop filters can save more than 15% of bitrate.

The competition platform is built on TAVS3 and Caffe. Participants only need to submit a Caffe model and its configuration file. Detailed information is available on the iQIYI AI Competition platform (click the original article link for the website). There is no fixed submission deadline; participants can continuously benchmark and verify performance. The working group will later establish a prize pool and award top teams. Participants are also encouraged to submit standard proposals to the AVS Working Group, where they can receive guidance from senior experts at the quarterly standard meetings, truly contributing to the standard‑making process and bringing research results into practical use. Best wishes for success in the competition.