[Feature] add Support for n_subsample

[jaschx]

Hi, it would be nice, if you can add Support for n_subsample to the Gui. (libvamf Option) Its speeds up Processing big Times, and from my personal Testing doesent lose to much Accuracy set to 5-10 .

wbr Alex

[fifonik]

Where can I read abot the option? In FFMPEG manual it is just mentioned, I have no idea what is the range of expected values.

P.S. right now you can adjust parameters supplied to ffmpeg in FFMetrics.conf.

[fifonik]

As I see, this is just a number greater then 0. 1 -- every frame calculated (default) 2 -- every 2nd frame calculated 3 -- every 3rd frame calculated and so on.

I've just tried with the param 1 and 5. Based on FFMpeg output the second one was just a bit faster on my PC with quite easy to decode m2ts & mp4 1080-60p files: 2.0x vs 1.4x. So If I understand correctly, the most time was in decoding, not in VMAF calculation. Is it really worth to have it in UI?

[jaschx]

It depends, on not superfast machines, like older I7s (4core 8 Threads) u get arround 3-4x speedup with 5 for example. Also, i use it with AV1, where decomp is slower overall. But Thanks for looking in to this.

[fifonik]

Also, i use it with AV1, where decomp is slower overall.

That was my point. On more advanced decoders like AV1 the decoding will take even more time so it will be smaller benefits from dropping off some frames during VMAF calculation. However, on bigger frames it should be more benefits.

P.S. I've already implemented it before testing performance gain. So now I'm not sure if I should merge it into main branch or not.

[jaschx]

Thanks, its up to you. On my older Maschine the Performance was going up times 4 with my 1080p Av1 Test Encode, thats why i was asking for it. Lets say it like this, its not hurting anybody to put it in (if you already did, why not ).

and a nice Sunday from Germany

[fifonik]

Will add it into the next version.

[fifonik]

Implemented in 1.5.0 beta 1 (just released).