Heh... yeah, those artifacts are a sign of the lossy compression the footage went through (I'm guessing xvid or rather low bitrate h.264, probably former taking into account the resolution), and there are some hints that it was already filtered to begin with. Anyway, aside from the aforementioned vinverse (which I assume you already tracked down and used, seeing as I don't see the interlacing leftovers anymore in these), I'm not really sure, there isn't much you can do without overdoing it at the same time, I think. The resolution is already low, so all the detail left is easily hit by now. Some things I did regardless were shifting the chroma (looks a bit shifted upwards, especially noticeable in the picture with the girl), dehaloing a bit (especially noticeable in the third image, around the red pillars... or whatever they are called, not sure), and a really slight debanding+addgrain to fix a bit the gradients messed up a bit by the quantizer.
I also cropped a bit the edge garbage and brought back to 640x360 (with h.264 mod16 isn't really as important as one might think, mod8 is fine).
So this would be what I came up with:
- Code: Select all
turnleft.chromashift(C=2).turnright
dehalo_alpha(darkstr=0.5)
crop(4,0,-2,-2).spline36resize(640,360)
gradfun2dbmod(1.005)
If you feel like it, you could also denoise a bit and perhaps sharpen too (2 of the images weren't really that bad as far as noise was concerned, but the third one has its issues, and I'm not sure what most of the source is like). In that case perhaps adding:
- Code: Select all
HQDN3D(3,3,0,0)
lsfmod(70)
after the crop and resize but before the debanding could be kinda ok but still somewhat detail killer. Yeah, I know, HQDN3D is pretty old and bad, but I needed something quick and lightweight that I could get to work just spatially and I couldn't think of anything else, sorry (I don't really think this is a case for a two stage fft3d... inb4 Hatt slaps me regardless). >_>