Questions about using Convolution3D

[aryan]

Well, let me be simple:

1. Where is right place to apply it - before or after cropping

2. What does "Movie Low Quality (noisy DVD source)" mean??? How to find which one is noisy and which is not (give me some example pls.)

3. There is presets for movie and anime. But which one to use for SHREK ? Generally its anime, but I think not exactly

4. Is there any sense of using GMC and PVE when using Convolution, and if yes can it produce some artefacts?

[hakko504]

I tend to put it after cropping, no need to denoise black borders that I cut anyway :) I'd say that most new films can use the HQ setting. I haven't done Shrek (yet), but most likely you can use MovieHQ for that one. It worked OK for Chicken Run :)

[Niño]

I always put convolution3d after croppig, like all the other filters (exept decomb) and before rsize, expecially with strong setting to prevent artefacts or excessive blurring.

[vlad59]

You should have a look a this 2 or 3 week old thread :
http://forum.doom9.org/showthread.php?s=&threadid=36344

[aryan]

Originally posted by vlad59
You should have a look a this 2 or 3 week old thread :
http://forum.doom9.org/showthread.php?s=&threadid=36344

Thanks for the advice ... really helpful info about resize and convolution :)

But how about using it with GMC and PVE? And is there any sense to use them at all if using Convolution3D ?

[Asmodian]

Do GMC and PVE have anything to do with convolution3D? I would think that they would be pretty much unrelated and it would be normal to use them together. Why wouldn't you want to use them when using convolution3D? Maybe I am missing something? :confused:

[JohnMK]

Quite unrelated. Use GMC + PVE + C3D if you want. I do . . .

[neo_sapien]

You probably shouldn't use c3d on Shrek. Shrek is a purely digital movie, there can't be much (any?) noise in it.

[JohnMK]

True, but there might be an appreciable increase in compressibility meaning he can use a higher resolution. I don't know if that results to a net quality increase, in fact, I'm betting the difference is negligible.

[vlad59]

Convolution3D works on a totally planar way. In the doc I only talks about pixel but I should talk about Y, U or V values so in the actual processing I simply don't care about the subsampling of chroma.

I know that's a pretty bad thing (but it seems to help compressibility). In a future version I'll try to allow only temporal, vertical, and horizontal filtering for chroma. But it's not for now.