A dependency-light Python script (standard library only) that runs VMAF on a (reference, distorted) pair with FFmpeg's libvmaf filter, then parses the per-frame JSON log and reports the score pooled four ways — mean, harmonic mean, 5th percentile, and the single worst frame — so you can see how much the headline mean hides. It prints the model used (so numbers are reproducible) and reads any gap against Netflix's ~6-point just-noticeable-difference rule of thumb. Includes the current FFmpeg/libv
Download free PDF
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
This Python script runs VMAF through FFmpeg's libvmaf and pools the score four ways, mean, harmonic mean, 5th percentile and worst frame, so you see what the headline mean hides. Our seven-volume handbook goes deeper: Volume 1 on VMAF and quality metrics, Volume 3 on bitrate ladders and QoE, and Volume 6 on AI models. Written by Nikolay Sapunov, CEO at Fora Soft.
Is your average VMAF score hiding bad frames in your encodes?
Send us a reference and a few encodes: we will measure them with proper pooling, find the weak scenes and propose ladder or encoder changes. 250+ video products shipped since 2005.
You get a realistic estimate, technical recommendations and next steps. Free, no obligations.
