Any way I can speed up my compression?

[TEKnician]

Is there a way to devote more RAM or something to increase compression speed? I'm barely getting 9fps with just

Code: Select all

ConvertToYV12()
tfm.tdecimate
maa()
gradfun2dbmod()
aWarpSharp2(depth=8)
crop(8,32,-8,-32)
LanczosResize(1280,720)

I notice that i'm only utilizing about 25% of my RAM. Any way I can allow more to get a faster result?

[Mister Hatt]

Use less inefficient shit for starters. Can't say it'll use much less ram but removing the converttoyv12 and putting your crop/resize directly after your IVTC helps. Do gradfun last and maa second last unless the warpsharp is causing aliasing. Some filters need to be run prior to crop or resize but none of the ones you have. Not that lanczos will give you aliasing in the first place fairly often so doing AA before it is dumb. Better parameters to tivtc might help.

Compression speed isn't really locked up by RAM anyway however your script rendering could be. I don't know what your original source is so I can't comment on that but you're probably handling it wrong. RAM usage at 25% generally indicates you're either not using all your RAM (ie not the bottleneck) or you're hitting a 32bit bottleneck in a 64bit system. Additionally, avs tends to only use 1 core so it'll most likely only peg one core, and that could be a slowdown. Profile your CPU cores to see their utilisation.

Not sure how much of that filter chain is required nor what encoder etc you're using but that might be something to re-evaluate.

[EvaFan]

Is your source HD and if it isn't is there a really good reason to upscale it that much?

The reason your probably wanting to use maa and warpsharp is probably because of what your doing to it by upscaling if anything, they would go after it.

If it is HD then I would just take awarpsharp and maa out cause your more than likely doing more harm than good, HD sources are very detail oriented you would probably be fine with just the deinterlacing, cropping, and resizing and maybe gradfun if the shaded/darker scenes have the banding like in the case of my HOTD blu-rays. Just my opinion anyway.

[EvaFan]

Also you can try adding Setmemorymax(amount) to the start of your script to see if your being limited on ram but I doubt that's the problem.

[TEKnician]

They're not HD, but I'm trying to learn ways of upscaling footage to a suitable quality. I'm mainly looking for ways to decrease the major defects like aliasing and banding.

Could my colorspace conversion cause the banding? If so, I can just remove gradfun2dbmod() from the filter chain.

[Mister Hatt]

Placeholder telling you to ignore EvaFan because he's usually wrong but I don't have time to read what he wrote write now or offer more advice. Give me a few hours.

[mirkosp]

Actually, I believe EvaFan's suggestion to have been spot on this time. It's mostly when he goes on about MT and 64bit avs that I don't quite agree with him.

[Mister Hatt]

Not really? Warpsharp is BETTER to use at high resolution where it can't cause too much damage, see supersampling and whatnot in contrasharpening or some of my AA processes. Same for anti-aliasing tbh. Also just saying, gradfun on the HOTD BD is one of the most damaging things you can do to it. I did kind of explain the RAM situation as well, but what it comes down to is maa() is a very inefficient filter, and warpsharpening after dithering is both inefficient and very damaging.

@TEKnician: Your colourspace conversion does nothing as the source is already YV12. You just need to move your filter chain around a bit. The other thing is that lanczosresize is a VERY bad resizer for upscaling. Firstly it causes serious ringing, especially when applied to details, which is my guess on why you thought you needed warpsharp. Next, it also is a rather sharp filter which probably made you think it was aliased. The banding is caused by oversmoothing it and then upscaling that. Pretty much your speed issues come from unnecessary filters and your quality issue comes from a shitty resizer. There are several ways of upscaling and the best that comes to mind is a method of splitting each RGB channel into a separate clip for filtering, and then merging them back together. The filter of choice there is a well tuned nnedi. There are a few other methods but I find that the be the best when it comes to undetailed chroma footage such as anime.

[EvaFan]

It's not that I'm always wrong, its just that u have to be always right and have the best suggestions hatter. You also forget there is no right or wrong way to use avisynth. As long as desired effects are achieved which is opinionative your answer might not even be the one the OP wants so deal with it bro.

[Mister Hatt]

The desired effect here is to speed it up. So by removing filters which are being used incorrectly and CAUSING the issues they are trying to fix, this end is achieved. I guess. Not so much that my input here is opinion, but it's certainly the best way to handle it instead of messing around with more useless crap. I also saw you offered no info on efficient upscaling.