Closed ls-milkyway closed 2 years ago
If you check the log after calculating VMAF metric using FFMetrics you should notice that n_threads is already in use. Currently, the program should put there a value that equal "a number of logical CPUs found in the system minus 1". This should be the best possible option in most cases. Note, the VMAF calculation still be much slower than PSNR/SSIM.
You can override the auto-detection logic by modifying FFMetrics.conf (usually not recommended, but could make sense if you run a few FFMetrics simultaneously or while doing some resource expensive tasks at the same time on the PC). Feel free to specify n_threads=1 and notice mush slower VMAF calculation speed.
You can also check #69
Nice app! ......specially the generation of graphs ....makes FFMetrics unique .....to increase VMAF calculation speed I could not find an option to assign CPU cores ....adding
n_threads=12
is all that you need to include in the ffmpeg command as stated here.