Log in

View Full Version : Help with video filtering


Cebez
1st August 2015, 18:01
Hello, guys^^

I'm relatively new to avisynth and its filters usage, so I was wondering if someone can tell me if I'm going too heavy with some filters or not.
I made some tries with an anime OAD:
http://screenshotcomparison.com/comparison/134690

The AVS script that I used is this:

FFVideoSource("videopath.mkv")
maa().maa()
GradFun3(lsb=true)
Dither_convey_yuv4xxp16_on_yvxx()
addgrainc(1,constant=true)


The third and fourth lines were suggested to me by a friend, he says that if I have a 8bit source and want to encode it in 10bit, I need a script like this:

FFVideosource("sourcepath.mkv")
GradFun3(lsb=true)
Dither_convey_yuv4xxp16_on_yvxx()

adding --input-depth 16 --input-res widthxheight to the cmd, but I really don't know if I can integrate it like that to the script or is better to encode a first time to filter and then a second time for the 10bit conversion.

Sorry if there are some grammar errors or if I didn't explain too well, but I'm not an native English speaker.

P.S. The screenshots were made with avspmod, but I don't know if screenshotcomparison uploaded them without quality loss. If it's better, I'll zip everything and upload it so something like mediafire.

colours
1st August 2015, 18:48
Rule 6! (http://forum.doom9.org/forum-rules.htm)

You should always encode from the source rather from another encode if you want maximum quality. It looks like most of the artifacts in your "source" aren't MPEG-2 artifacts so encoding directly from the DVD would already look much better even without filtering.

Also, AddGrainC is an 8-bit filter and you shouldn't be using it when your clip has 16-bit content. There's Dither_addgrain16 for that.

Keiyakusha
2nd August 2015, 17:41
Cebez
Not really sure why would you want to use maa() not to mention 2 times in a row. It looks like it does makes lines look a bit better, but in general it is not made to be used in this case. Maybe you can try some deringing instead or some kind of masked smoothing like dfttest, which is also a good deblocker (it is not masked without additional scripting but may be useful even bare).
3rd and 4th line make sense to me, addgrainc is not. What exactly you're trying to achieve by adding it? Except increasing filesize? While what colours said about 8 vs 16 bit is true, I personally don't see the reason for this filter to be present at all.
Anyway, I agree that the source you got doesn't looks like something worth to be wasting time on. And If I was you, I wouldn't be dealing with upscales. There are little reasons not to keep original DVD resolution or making it something like 852x480, if you absolutely can't have anamorphic encode.