Gracias, Gracias, Gracias, Thanks, Thanks X 1000

-kwag

hehehe you're both welcome .

I just uploaded to my site version 0.7, which has a new method: dithering. It's very similar to the noise method except it will add the same noise to each frame, creating a kind of "unchanging noise" effect.

I'm hoping that this will eliminate the artificial "movement" created in static parts of the frame by the noise method. Theoretically this might produce a more natural-looking result and might even be a little more compressible.

Let me know if it works .

Man, these filters are wonderful. What should be added to our AVS scripts to take full advantage of the new dithering feature in 0.7? I see the usage rules in the readme, but does anyone have some recommended parameters?

By the way, I've been using the KDVD *HALF* template and applying the fluxsmooth 0.4 and blockbuster 0.6 filters... FLAWLESS video. I cranked the CQ up to 80 (just cause) and I can't tell the difference between the source file and the one that takes up a quarter of the space! I'm not worthy!

Kwag,

Thanks for all your work, and just a quick question re your script that you posted for the LBR template. Should I change the BB_Resolution and BB_StrengthConstant lines to reflect PAL source (ie. 352*28, and what changes would need to be made to the script if I want to use the kvcd3 templates instead.
I am capturing at 704*576.

You'd use it in a very similar way to the "noise" method. The only difference will be that the noise added will be the same for each frame, instead of changing. Try it like this:

Blockbuster(method="dither", min_detail=1, max_detail=100, variance=50)

and it should look like you're watching the clip through a speckled window.

As for what you can do with it: test it! Let me know how it compares, in the final encode, to using method="noise" with the same parameters and with different parameters.

As I said in the docs, it's experimental at the moment. So experiment .

Maybe it works a little better like this: Blockbuster(method="dither", detail_min=1, detail_max=100, variance=50)
( Easy on the beer SansGrip, it's not time yet )

-kwag

Er, what he said .

Will do! I'm going to encode some samples once I get home from work. I'm mostly interested in the extra compression-- if there is any-- since I really can't imagine the video looking any better than it already does!

My very brief test indicated there is some extra, at least with the source material I used. With method="dither" on a 100-second clip I got something like 11.4mb and with method="noise" I got something like 11.6mb. Not a whole lot, but it adds up.

Mind you, I didn't do a visual compare to see if the decrease in size was caused by an increase in blockiness, so YMMV .

Looking good
I'm testing it like this: Blockbuster(method="dither", detail_min=1, detail_max=10, variance=.7)
No need to dither high frequency components ( At least that's my thinking ).
So just dither the areas where DCT blocks are present. Low lit, etc.

-kwag

Yep. The detail_min=1 detail_max=100 was just to demonstrate the difference between method="noise" and method="dither".

Arghh! Curse you SansGrip! The last thing I need is more choices...

At best, in my opinion, dither just looks different from noise. I guess it's just a matter of preference, eh? I would imagine some sources will look better when encoded with dither instead of noise, but in my case I couldn't tell much difference.

Files are consistantly smaller with dither, but only by a few hundred kilobytes. As mentioned already, I'm using Kwags Half KDVD template. Source files are DVD rips.

Keep up the hard work, guys! This stuff is amazing! Patent pending?

I know just how you feel .

And that's really all it is: a bit different from noise .

My hope was/is that with the right variance setting it'll reduce the artificial "motion" I sometimes see on walls etc., with DCT blocks kinda jumping around. This occurs, I think, when the bit rate isn't quite high enough to describe the noise fully. Theoretically dither should stop that, since the noise never changes from frame to frame. A static wall should always have the same pattern of noise on it, and so the DCT blocks shouldn't jump.

That's the theory anyway. We'll see if it works out in reality .

@SansGrip,

I tested Blockbuster "dither" using 528x480 CQ version.

Blockbuster(method="dither", detail_min=1, detail_max=10, variance=.7)

I didn't notice any effect on the walls or any static background areas
having a more stable appearance. Maybe Tmpgenc's CQ ignors this
frequency. I increased detail_min and detail_max. It still looked the
same. Have you made any tests that showed a differance.

-black prince

I think that noise and CQ doesn't like each other...

It does, unless the CQ is very high (>80 or so). I'd say Blockbuster is most useful with 352x240 non-LBR and 352x480 resolutions, using CQ_VBR. I have better results not using it with the LBR template, because by definition you need all the bits you can get just to encode the un-noised picture...