JVET-AM0298 Ultra-low latency demo: Ressource constraint VVC and HEVC software encoding up to 4K [M. Alvarez Mesa, B. Bross (HHI), C. C. Chi (Spin Digital Labs)] [late]
This document describes a demo setup for ultra-low latency use cases based on HEVC and VVC software encoding. Two synchronised Spin Digital Labs software encoders (spin enc live) are encoding a webcam video live stream at low bit rates on a Mini-PC with a 16 core AMD Ryzen 9 9950x. One spin enc live instance encodes HEVC and the other one encodes VVC video, and both are sent as two synchronised streams in an MPEG-2 transport stream. The input video can be changed from 1080p60 to 2160p30, and each stream can be set to various low bitrates from 200 to 500 kbit/s constant bit rate (CBR). Both encoders are restricted to use 7 cores (14 threads) exclusively. It is shown that the VVC video quality is clearly better than the HEVC video quality for the same low bit rate when both encoders are given the same amount of computing resources. In addition, the use of Gradual Decoder Refresh (GDR) in both HEVC and VVC significantly improves the video quality by allowing the higher rate intra blocks to spread across multiple frames. The demo highlights the usability of VVC for ultra-low latency use cases in resource constraint environments.
Demo setup diagram:
HEVC and VVC encoding settings:
Settings | HEVC | VVC |
Codecs | HEVC Main 10 | VVC Main 10 |
Rate control | CBR | CBR |
Bitrate | 500 Kbit/s | 500 Kbit/s |
Coding structure | GOP 1 (Low-delay P) | GOP 1 (Low-delay B) |
Lookahead frames | 1 | 1 |
Frames in flight | 1 | 1 |
Intra period | 60 frames | 60 frames |
Gradual refresh (GDR) period | 30 frames | 30 frames |
HRD buffer | 1000 ms | 1000 ms |
Preset | balanced | balanced |
CPU cores (threads) | 7 CPU cores (14 threads) | 7 CPU cores (14 threads) |
Streaming | MPEG2-TS - RTP | MPEG2-TS - RTP |
End-to-end latency about 120ms.
Complexity of HEVC and VVC software encoders on par.
One subjective comparison, 4Kp30 encoding at 500 Kbps, HEVC (left) vs VVC (right), is shown below in the case of close-to-camera movement. Clear subjective quality benefits are observed.
The demo uses the GDR feature in HEVC and VVC, though some implementation details are slightly different.
It was suggested that low-latency, low complexity VVC encoding is quite achievable for the use case presented.
Profile/tier/level specification (0)
This section is kept as a template for future use.