Understanding VMAF
VMAF, developed by Netflix, is a perceptual video quality metric that evaluates the visual quality of encoded videos based on human perception. Unlike traditional metrics such as PSNR (Peak Signal-to-Noise Ratio) or SSIM (Structural Similarity Index), which focus solely on pixel-level differences, VMAF takes into account various factors that influence visual quality, including spatial and temporal characteristics, color accuracy, and motion smoothness.
Calculating VMAF
VMAF is calculated using a machine learning model trained on human subjective ratings of video quality. The model predicts how humans would perceive the quality of a video based on its encoding parameters and content characteristics. VMAF scores typically range from 0 to 100, with higher scores indicating better perceived video quality.
The calculation of VMAF involves several steps:
- Feature Extraction: Relevant features such as luminance, contrast, and spatial frequency are extracted from the reference (original) and distorted (encoded) videos.
- Feature Transformation: The extracted features are transformed using a set of quality models that simulate human perception.
- Feature Fusion: The transformed features are combined using a fusion model to generate a final quality score, representing the perceived video quality.
How we reached our VMAF ideal quality settings:
Our journey to identifying the ideal VMAF quality settings has been a result of meticulous testing and analysis conducted over an extensive period. Through this rigorous process, we've determined that the 90%, 93%, and 96% VMAF settings strike a perfect balance between visual quality and compression efficiency.
To achieve these desired VMAF values, we encoded the Original Video to AV1 using carefully crafted parameters defined by our team. Testing various parameters through trial and error, we arrived at three distinct levels of quality:
Premium: 96% VMAF
Optimal: 93% VMAF
Standard: 90% VMAF
Conducting numerous tests, we iteratively adjusted the settings until we ultimately attained these values and quality levels.
The diagram and the instructions below explain our methodology:
Utilizing the following parameters with ffmpeg, we executed the VMAF process:
ffmpeg -i original.mp4 -i output.mp4 -filter_complex "[0:v]scale=3840:2160:flags=bicubic[main];[1:v]scale=3840:2160:flags=bicubic[second];[main][second]libvmaf" -f null -
This command enabled us to generate a comprehensive table and visualization showcasing the calculated VMAF scores, as seen in the accompanying image.
These results underscore the exceptional compression efficiency of our encoding solution powered by VMAF, allowing for significant file size reductions while maintaining a consistent high VMAF score.
Case Study: Encoding a video with SlashedCloud:
In a real-life scenario, we acquired a video sample that was encoded in H.264 with a resolution of 4K. Leveraging our advanced encoding tool, we seamlessly transcoded the file to AV1 format, achieving remarkable results using our three quality settings:
- 4K
Original H.264 Size: 1.6 GB
AV1 Encoded Size: 207.66 MB with VMAF: 96.45% - (Premium Quality)
AV1 Encoded Size: 85 MB with VMAF: 93.06% - (Optimal Quality)
AV1 Encoded Size: 56 MB with VMAF: 90.07% - (Standard Quality)
Here’s an example of the original vs the encoded:
(Original)
(Encoded)
Our premium quality setting, boasting a VMAF score of 96.45% and an encoded size of 207.66 MB, provides unparalleled visual fidelity suitable for high-end applications where pristine quality is paramount. This setting achieves a compression size of approximately 87%, demonstrating a substantial reduction in file size compared to the original 1.6 GB.
The optimal quality setting, with a VMAF score of 93.06% and an encoded size of 85 MB, strikes an excellent balance between quality and file size reduction. This setting achieves a compression size of approximately 94.7%, catering to a wide range of viewing environments and devices.
Lastly, the standard quality setting, with a VMAF score of 90.07% and an encoded size of 56 MB, offers a compelling option for scenarios where conserving bandwidth and storage space is essential, without compromising on perceptible visual quality. This setting achieves a compression size of approximately 96.5%.
These results underscore the versatility and effectiveness of our advanced encoding tool, which empowers users to tailor their encoding preferences according to their specific requirements, while ensuring an optimal viewing experience across diverse platforms and resolutions."
Conclusion
SlashedCloud is committed to delivering the highest standards of video quality for our clients. By leveraging VMAF and advanced machine learning algorithms, we ensure that every video optimized through our platform meets the rigorous demands of today's digital landscape. Experience the difference with SlashedCloud and elevate your content to new heights of visual excellence.
