NVIDIA has released the latest iteration of its Video Codec SDK, version 12.2, which brings significant improvements to video quality, particularly for high-efficiency video coding (HEVC). According to the NVIDIA Technical Blog, the update offers a comprehensive range of APIs for hardware-accelerated video encoding and decoding on both Windows and Linux platforms.
Key Features of Video Codec SDK 12.2
The Video Codec SDK 12.2 introduces several new features aimed at enhancing video quality and reducing bit rates. These features include:
- Lookahead Level: This feature increases lookahead analysis to optimize coding efficiency, particularly for HEVC. It allows for better bit allocation across frames.
- Temporal Filtering: Designed to filter out noise and improve compression efficiency, this feature uses motion estimation to reduce the bit rate required for encoding natural video content.
- Higher Bit-Depth Encoding: This feature enables the encoding of 8-bit content as 10-bit, improving correlation and resulting in better compression for both HEVC and AV1.
- Unidirectional B-Frames: These frames improve quality by using predictions only from past frames, making them useful for low-latency encoding scenarios.
- Ultra High Quality (UHQ) Tuning Info: This feature provides the best quality for latency-tolerant use cases by combining lookahead level and temporal filtering optimally.
Detailed Feature Insights
Lookahead Level: This feature uses coding tree units (CTUs) and other encoding statistics to improve rate control. It offers four different lookahead levels, each with varying trade-offs between performance and quality. Settings include enableLookahead to turn on the feature, LookaheadDepth to determine the number of frames buffered for analysis, and LookaheadLevel to set the level of analysis.
Temporal Filtering: This feature reduces noise in natural video content by using patches from neighboring frames. Implemented using a CUDA kernel for filtering and the NVIDIA encoder (NVENC) hardware for motion estimation, temporal filtering offers an average coding gain of 4-5% for natural video content.
Higher Bit-Depth Encoding: By encoding 8-bit content as 10-bit, this feature provides coding efficiency gains of 3-4%. The conversion happens within the NVENC driver using CUDA for HEVC and hardware for AV1, with negligible performance penalties.
Unidirectional B-Frames: These frames predict from the two previous frames, offering 3-4% BD-BR savings over P-frames. This feature is particularly useful for low-latency encoding and can be enabled by setting enableUniDirectionalB to 1.
UHQ Tuning Info: This feature combines lookahead level and temporal filtering to provide optimal settings for various presets, especially for latency-tolerant encoding. It fixes the number of B-frames to five and disables adaptive I and B-frames, using a fixed GOP structure.
Getting Started with Video Codec SDK 12.2
The Video Codec SDK 12.2 offers significant improvements in video encoding quality for HEVC, with new features designed to reduce bit rates and enhance overall performance. Users can download Video Codec SDK 12.2 to get started and join discussions in the NVIDIA Developer forums. For more insights into the new capabilities, see the NVIDIA Technical Blog.
Image source: Shutterstock