Advanced Video Encoding Techniques in Tencent Meeting: Temporal SVC, ROI‑Based Encoding, Screen Content Coding, and YUV444 Support

Tencent Meeting is a real‑time audio‑video communication product that must transmit massive video data such as camera streams and screen‑share content. Because of the large data volume, the video is first encoded and compressed before transmission.

In video coding there are three frame types: I‑frames (intra‑predicted, independently decodable), P‑frames (forward‑predicted, depend on previous frames), and B‑frames (bidirectional‑predicted). For low‑latency real‑time communication B‑frames are usually avoided, leaving only I‑ and P‑frames.

The classic IPPP structure (one I‑frame followed by several P‑frames) provides good compression but is vulnerable to packet loss: if a P‑frame is lost, all subsequent frames that reference it cannot be decoded, causing long stalls. To mitigate this, Tencent Meeting adopts a temporal Scalable Video Coding (SVC) scheme that divides frames into hierarchical layers (Layer0, Layer1, Layer2). Lower layers can be referenced by higher layers, but not vice‑versa, so loss of a higher‑layer frame does not affect lower‑layer decoding.

Another optimization is ROI (Region‑of‑Interest) detection and ROI‑based encoding for camera streams. A fast face‑detection algorithm identifies the facial region, which is then allocated a higher bitrate while the background receives a lower bitrate. This approach improves subjective quality at low bitrates and saves 20‑30% bitrate at high bitrates.

For screen‑share and whiteboard scenarios, traditional block‑based transform coding (H.264/H.265) loses high‑frequency details. Tencent Meeting therefore incorporates Screen Content Coding (SCC) techniques such as intra‑block copy (using motion vectors in I‑frames) and palette mode, which encodes pixels by transmitting palette indices instead of transformed coefficients. These methods greatly improve compression efficiency and visual quality for screen content.

Because screen content often contains rich chroma information, the system also adds YUV444 support. While most video applications use YUV420 to reduce data size, YUV420 severely degrades color fidelity for screen sharing. Experiments show that YUV444 preserves color detail and avoids the gray‑scale artifacts seen with YUV420.

Finally, Tencent’s proprietary H.264 and H.265 encoders have been heavily optimized. Compared with the open‑source x264/x265 encoders, the Tencent H.264 encoder achieves >40% compression gain with only ~11% speed loss for screen content, while the H.265 encoder delivers up to 83.7% compression improvement and 210% speed increase in screen‑share scenarios.

The article concludes that these encoding “artifacts” continuously evolve to enhance user experience across different video scenarios in Tencent Meeting.