Trying to find the cause of a huge flaw in a filterchain...

[Backwoods]

Originally posted by Chainmax
One more thing, I don't understand what you mean when you say that overscan(?) makes cropping needless, but aWarpSharp and aWarpSharpDering might cause some green or black splotches on the borders, plus there is a bit of noise and haloing around the borders, so just crop accordingly and use 720,480 as parameters in LimitedSharpen. IMO it's better that way.

Ok, I didn't think green or black would come up from using that filter. I didn't bother cropping because TV overscan will hide those edges.

I'll try the latest script you posted tomorrow or so when I capture another anime VHS tape. And of course I will post samples.

[tritical]

P.S: tritical, is TDeint's link=1 better than the other link modes? Sorry for the late response, but really no mode is better then the other... it simply depends on your objective. Link = 1 will link all three planes Y, U, and V. That is, after static area detection, if an area in any one plane is detected as moving the spatially corresponding parts of the other two planes will also be marked as moving if they weren't already detected as moving. Link=2 only does Y->U/V so for any moving parts detected in Y the corresponding parts in the U and V planes will be marked as moving (that is the default value in the latest version). Link=3 does U/V->Y (pretty useless) and link=0 does no linking at all. Link=1 will leave the least residual combing at fixed lthresh/mthresh settings, but is overkill for most material and may cause unneeded interpolating of pixels in the luma plane. Link=2 is probably the best compromise and general purpose setting. However, link=0 is usually fine to use if your motion thresholds are relatively low.

[Chainmax]

I see, thanks for the answer :).

[Backwoods]

I've compared your previous script to the last one you posted and it seems the previous one does a better job. There is another old anime I captured and I can post samples tomorrow (hopefully), also I got LRemoveDust to work thanks for your help.

plus clear out some more PMs

[Chainmax]

It seems weird that the difference should be too noticeable since I only replaced RemoveGrain(1,2,2) for Convolution3d(preset=animehq") and went back to LimitedSharpen's default of Strength=100. How exactly do you feel the last script is worse?

P.S: I don't understand why the Inbox reacts as full as I only have about 15 messages on it. I will delete all messages and clear the Sent box a bit too.

[Backwoods]

Original Capture
http://www.uploadhouse.com/images/797818631ULTRA_ORG.JPG (http://www.uploadhouse.com/)

Previous Script
http://www.uploadhouse.com/images/667947784ULTRA_NEW_01.JPG (http://www.uploadhouse.com/)

Latest Script
http://www.uploadhouse.com/images/852775363ULTRA_NEW_02.JPG (http://www.uploadhouse.com/)

[Chainmax]

I see almost no difference (maybe the last script has a bit more detail?), but then again my eyesight isn't all that trained and mi monitor is FUBAR. By the way, ChronoCross is helping me out in another clip and pointed out some nasty artifacting due to the aWarpSharping. Therefore, the script (wether it's the old one or the new one) must be altered. I hope to post a fix soon.


P.S: my Inbox should be cleaned by now. Also, I really think you should crop and use 720,480 as resolution parameters in LimitedSharpen.

[Backwoods]

Originally posted by Chainmax
P.S: my Inbox should be cleaned by now. Also, I really think you should crop and use 720,480 as resolution parameters in LimitedSharpen.

Ok, I'll resend the PM.

I do use 720x480 in LS but like I said before I'm not cropping because of TV overscan. The sides won't show up as is. I guess I could mask the edges so there isn't any tearing.

A good way to tell the difference between two pictures is to save them to your HDD and flip back and forth.

[Chainmax]

Sure, that works when using a standard TV, but what about watching it on an HDTV or a computer monitor? Anyway, wait for the next version of the script because like I pointed out, there's a serious issue concerning aWarpSharp.