BILIVVC Secures Third Place in 2022 MSU Encoder Competition (1080p 1fps & 5fps)

On July 14, 2023, the results of the 2022 MSU Encoder Competition were announced. Bilibili’s self‑developed VVC encoder, BILIVVC, participated for the first time and achieved a comprehensive YUV‑SSIM performance ranking of third place in both the 1080p 1 fps and 5 fps tracks.

The MSU Encoder Competition, organized by Moscow State University, is a world‑class contest that attracts leading video‑codec vendors. The competition’s Full‑HD track evaluates objective quality metrics such as SSIM, PSNR, and VMAF across a diverse set of test sequences that consider temporal‑spatial complexity and bitrate coverage.

Figure 1 shows the Full‑HD results of the 2022 competition, and Figure 2 highlights the performance of BILIVVC (indicated by the red arrow) in the 1080p 1 fps track.

BILIVVC is a pre‑research encoder project of Bilibili’s video‑encoding team. Initiated at the end of 2021, the encoder was delivered to the MSU competition in under one year while remaining a small‑team effort, making its third‑place achievement noteworthy.

Built on the H266 core, BILIVVC implements the vast majority of VVC standard tools and applies extensive optimizations. Compared with the reference implementation, BILIVVC’s tool implementations are more efficient, and dozens of fast algorithms have been devised to boost encoding speed with minimal quality loss. The encoder supports multiple rate‑control modes, including CRF, VBR, and two‑pass, facilitating future deployment on Bilibili’s platform.

Beyond the core codec, BILIVVC incorporates a high‑performance pre‑analysis module that leverages cutree and adaptive quantization (AQ) techniques, further refined for the new VVC standard. The team also developed scene‑switching and adaptive frame‑structure algorithms tailored to VVC, improving performance across varied content. Optimizations to the motion‑compensated temporal filtering (MCTF) based on cutree analysis yielded over a 5 % gain compared to the VTM reference, while adjustments to frame‑structure and LMCS (Luma Mapping with Chroma Scaling) contributed an additional 2 % improvement.

Given VVC’s roughly 50 % compression efficiency increase over HEVC, its computational complexity is significantly higher. The BILIVVC team re‑engineered data structures for new tools such as dependent quantization and the Adaptive Loop Filter (ALF), applying assembly‑level optimizations that dramatically increase execution speed relative to the reference code.

In terms of fast‑algorithm design, BILIVVC emphasizes inter‑tool cooperation. By exploiting the characteristics of newly added VVC tools, the encoder allows predictions from multiple tools to inform each other’s fast‑path decisions, achieving substantial speed gains with negligible quality degradation.

The Bilibili Multimedia Lab, a technology‑driven young team, focuses on delivering crystal‑clear video experiences. Its portfolio includes controllable‑bitrate narrow‑band HD transcoding, lossless visual pre‑processing, ultra‑real‑time 4K 60 fps super‑resolution live streaming, and an efficient video‑image processing engine, all of which enhance the overall performance and cost‑effectiveness of Bilibili’s video ecosystem.