Netflix VMAF is an Emmy-winning perceptual video quality assessment system that fuses multiple feature extractors to produce human-correlated quality scores. It ships as a C library (libvmaf), a Python wrapper with tools for training and validation of custom models, and a command-line vmaf tool that also reports PSNR, PSNR-HVS, SSIM, MS-SSIM and CIEDE2000. Integration points include an FFmpeg filter, a Docker image, and Windows build instructions. The package includes documentation, model selection guidance, and utilities for dataset processing and visualization, enabling deployment in production and research workflows.
Development emphasis has been on accuracy, extensibility and speed. Recent milestones include the release of new VMAF models (v1), libvmaf v3.0.0 with bug fixes and API cleanups, and earlier SIMD-optimized fixed-point implementations (AVX2/AVX-512) that doubled performance. Quality-control features introduced include a No Enhancement Gain (NEG) mode to avoid rewarding artificial enhancements and the open-sourced CAMBI banding detector with full-reference and contrast tuning options. Licensing moved to BSD+Patent, AOM adopted VMAF for CTC, and contribution and build guides are provided for extending models and feature extractors.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
![IMF World Economic Outlook 2026 [pdf]](https://hntoday.pixelshot.io/screenshots/20261019/fef3e7cddab04fea_tn.jpeg)
The IMF World Economic Outlook 2026 provides projections and analysis of global economic trends for the year 2026. It covers topics such as economic growth, inflation, and financial stability across different countries and regions. (imf.org)
Astronomers have discovered a new green star that exhibits unusual characteristics. The star's unique properties have captivated the scientific community and prompted further study. (reuters.com)
A crisis in mathematics is highlighted by Terry Tao, who calls for urgent efforts to address ongoing challenges in the field. The article explores the difficulties faced by mathematicians and possible solutions to improve the discipline. (nytimes.com)
Large language models are proving capable of generating mathematical proofs and accelerating scientific discovery. However, their integration must be carefully managed to prevent a future where human understanding diminishes and knowledge becomes concentrated and opaque. (terrytao.wordpress.com)
Neuroscientist Jared Cooney Horvath states that Gen Z is the first generation in modern history to score lower on standardized tests than the previous one, across multiple cognitive areas. He attributes this decline to increased screen time, which reduces attention and social interaction crucial for learning. (vice.com)
Researchers demonstrated that environmental memory in non-Markovian systems can be harnessed to extract work, surpassing energy stored in observable degrees of freedom. Using time-delayed measurements on an optically trapped particle, they showed that information backflow from hidden bath degrees of freedom enhances work extraction, revealing environmental memory as a thermodynamic resource. (nature.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.