View Full Version : Faster with filters?


Xodarap
18th October 2011, 04:21
(This is really an avisynth thread, but since I'm using megui, I figured I'd post it here)

This may be the result of fiddling around with SetMTMode, which is new to me. The "problem" (which I'm assuming is a problem because it seems wrong to me) is that I'm getting significantly faster encoding *with* TemporalSoften included than without.

Here's the rundown:

Script:
SetMemoryMax(1536)
SetMTMode(2,0)
FFVideoSource("...", threads=1).AssumeFPS(24000,1001)
TemporalSoften(2,2,2,5,2)
LanczosResize(720,400)

= 52.8 fps

If I remove the TemporalSoften line, I get about 33 fps. Not a small difference! Can someone explain to me why this is happening? Is this normal, or am I doing something wrong?

LoRd_MuldeR
18th October 2011, 13:54
Are your FPS number from the plain AVS script -or- do you get these numbers when encoding from the script?

In the latter case, it is quite possible, that the "soften" filter smoothes out more details beforehand and thus makes the video easier to compress.

So you might actually safe the time in the encoding stage - at the cost of reduced details ;)

Xodarap
18th October 2011, 17:16
Yeah, it's encoding fps. And I that makes sense; especially since the temporalsoften filter does kill a fair bit of detail. I assumed, though, that the process of temporal blending would still have to take longer than the encoding speed boost it could give. I never noticed this before.

LoRd_MuldeR
18th October 2011, 17:46
I assumed, though, that the process of temporal blending would still have to take longer than the encoding speed boost it could give

Nope, the TemporalSoften filter is more a "quick and dirty" denoising filter. It is extremely fast, compared to more sophisticated denoise filters, e.g. such ones that use motion detection. At the same time x264 is a very complex/sophisticated but also highly-optimized encoder. It uses "early termination" in many of its algorithms, which explains why speed can vary significantly with the same settings, depending on the input.