New luma correction in XviD

[molerus]

Hi!

I recently encoded "The Terminator" into 2 CD's using the filter called "PixieDust" and "Convolution3D". As I might expect the large areas with uniform color with high motion (strob light in the disco), as well as smoke look TERRIBLY SHITTY (lack of noise caused catastrophical blockiness). On the other hand the rest of the muvie looks just perfect.:p
As far as I know XviD doesn't do any luma correction in curve scalling as NanDub did. I'm about to write the AviSynth filter which would store luma information to an auxilliary file during the first pass, and then would mix it into stats file. I am planning to use GordianKnot to scale the curve, using the luma information.

My question is: 1) can anyone give me a hint how the stats file is built (link maybe), 2) how to asses the luma of frames (uniformness of colors), perhaps NanDub manner????

BTW. Does XviD correct the curve scalling using the motion information, and if how strong the setting are? Is motion calculated using the information from motion estimation, or is it just the absolute difference between the consecutive frames (in chroma for example). And another thing: why the values of motion when loaded into GKnot are so high (almost the hole movie with 100% motion)??? Is it assesed similary to NanDub or the values are different??

[Didée]

For your technical questions, better wait for someone to jump in who lives, sleeps and eats in the deeps of the codec ... I could answer some questions, but with danger of misinformation :)

Anyway, I wonder how you managed to make Terminator look shitty on 2 CD's - okay, not the whole movie, just some scenes.
I would NOT encourage to use C3D together with dust, especially not for a 2CD-rip. Why filter the hell out of it when you have enough bits available? With that combo you used, you'd almost reach a maxed-out encoding.
I would use one of those filters *alone*, and throw in some small amount of noise again afterwards with blockbuster. ( :confused: ? That's serious, really: first filter out the *general* noise, then put some noise back to only those areas that *need* noise, against blocking)
Also, be aware that ATM dust may produce some blocking artefacts of its own - namely in flat areas, walls and smoke! This pretty well could be the problem you're facing.
So load your script again in Vdub, and look at exactly those scenes that came out blocky after conversion. To see things better, give some additional sharpen at the end of your script (or in Vdub directly).
Dust is very promising, but far from perfect yet - alas.

Regards

Didée

[Teegedeck]

Originally posted by molerus

As far as I know XviD doesn't do any luma correction in curve scalling as NanDub did.

The abovementioned feature isn't neccessary for XviD, it was only there in Nandub to compensate for DivX;-)' mistakes.

Don't filter too much, one noise-reducing filter would suffice - or use blockbuster after filtering, as mentioned by Didée.

Does XviD correct the curve scalling using the motion information, and if how strong the setting are?
No, it doesn't. Motion-based curve-scaling was a bad idea in Nandub, already. Perhaps we'll see a block-based removal of high requencies or stronger quantization for fast-moving objects in XviD one day, as part of some improved adaptive quantization. That would be a meaningful use of human visual perception regarding motion. But it will not be tomorrow.