Mezzanine Codec Quality Benchmarks

Executive Summary

A benchmark compared NDI SpeedHQ and JPEG XS as mezzanine codecs used in production workflows, focusing on image fidelity, bitrate, and generational quality loss. The tests were run on an Amazon EC2 C6id.12xlarge instance and evaluated JPEG XS across compression ratios from 7:1 to 15:1. Results included a finding that both codecs exhibited less than 0.5 dB PSNR quality loss over subsequent generations.

Key Industry Developments

• Independent benchmarking was performed to compare NDI SpeedHQ and JPEG XS on image fidelity, bitrate, and generational quality loss, framing both as codecs used in production workflows.
• The benchmark execution environment was specified as an Amazon Elastic Compute Cloud (Amazon EC2) C6id.12xlarge instance, anchoring the measurements to a defined compute configuration.
• JPEG XS encoder settings were varied using compression ratios from 7:1 to 15:1, indicating a parameter sweep across multiple operating points.
• Generational quality behavior was quantified using PSNR, with both codecs showing less than 0.5 dB PSNR loss over subsequent generations in the tested setup.

Real-World Use Cases

• Live cloud production environments can use benchmark data on image fidelity, bitrate, and generational quality loss to inform selection between NDI SpeedHQ and JPEG XS as mezzanine codecs in production workflows.
• Workflows that involve long signal chains (modeled as repeated encode–decode cycles) can use the reported generational PSNR behavior to estimate cumulative quality impact when using either codec.
• Engineering teams evaluating JPEG XS operating points can map compression-ratio choices (7:1 to 15:1) to measured outcomes from the benchmark when configuring contribution or processing pipelines.

Why It Matters

Mezzanine codecs are commonly used inside production workflows, so measured tradeoffs among image fidelity, bitrate, and generational quality loss directly affect end-to-end quality in processing chains. A defined test environment (Amazon EC2 C6id.12xlarge) helps teams understand the compute context in which results were produced and supports more controlled comparisons when reproducing tests. Reporting that both codecs showed less than 0.5 dB PSNR loss over subsequent generations provides a concrete indicator for scenarios where content may be encoded and decoded multiple times.

Sources

• https://aws.amazon.com/blogs/media/benchmarking-video-codec-quality-by-comparing-ndi-speedhq-and-jpeg-xs/