Script for plotting the output of easyVmaf (https://github.com/gdavila/easyVmaf)
Initial code from https://github.com/master-of-zen/Plot_Vmaf
Further code from https://github.com/PhilippeR/Pra_Plot_Vmaf
Requirements: easyVmaf for doing VMAF measures ffmpeg for extracting frames Python3 with numpy, matplotlib, Pillow
This version improves on :
- Being agnostic for the distinctive field for 4k and hd in the EasyVmaf output.
- Using the Mean and Harmonic mean values from the EasyVmaf JSON output instead of recalculating.
- Introducing .50 and .99 percentile in the histogram.
- Creating a a more useful grid on the plots, so it's easier to read the JND of 5 VMAF points.
- Focussing on Harmonic Mean instead of Mean values as indicator in the plots.
- Option to export frames with a VMAF score below the 0.01 percentile as TIFF files. NOTE: Please use the TIFF export conciously and with care. Using this on a large video file can cause a multitude of TIFFs being exported.
python plot_vmaf_improved.py [-h] [-o OUTPUT] [-p] vmaf_file1 vmaf_file2 ..... vmaf_fileX python plot_vmaf_improved.py vmaf.json -o plot.svg-o --output ["file"] name and format of the temporal VMAF graph (default plot.png)
-p generate percentile graph
-f export frames with vmaf below 0.01 percentile to TIFF
According Zhi Li (https://netflixtechblog.com/vmaf-the-journey-continues-44b51ee9ed12) and Jan Ozer (https://streaminglearningcenter.com/blogs/compute-vmaf-using-ffmpeg-on-windows.html) , harmonic mean makes more sense than arithmetic mean.