g-force

Mielr said:

dark_jedi says that the 'gold trim' problem in SW has reappeared with the latest tweak (is this V. 5 or 6? I've totally lost track. LOL.) Is there a simple way to fix this without affecting other scenes (like you did on ESB)?

No, no simple fix. They are really very different issues. The ESB problem was caused by something I was doing, the SW "gold trim" thing is in the source. The latest version of the script corrects all the artifacts that the motion compensation caused. Unfortunately, there are a lot of things that the motion compensation helped (like the 'gold trim'), but in repairing the motion compensation, a lot of the problems with the source come back.

I haven't figured out a way to have it both ways, so I'm erring on the side of "do no harm".

-G

zee944 said:

Oh well. I thought my post will be useful, but not at all.

On the contrary, You got me thinking about de-haloing and de-ringing, and how nobody (in general) has done this correctly yet. It turns out that it's really a deconvolution problem, and that is proving difficult but seems do-able.

-G

V.4 just kills too much FINE detail.  Yeah, it sharpens medium to low frequency detail, but just kills all the fine stuff. It also has too much artifacting on quickly moving objects.

Might want to wait until I have my de-ringing script ready. It's looking good so far, just want to be sure before I make a fool of myself (yet again)

But, that'll look more like V.6 anyway. If you want to fix V.4, just replace the entire "global motion" stage with the new one.

-G

Okay, after seeing Red5's screen caps, I retract all previous statements. The aliasing is in the master tape then, if it's in both the GOUT and the X_ captures.

Shoot, I really thought I was on to something. I was under the impression that there was much less aliasing in the laserdiscs, and if the aliasing was a symptom of bob de-interlacing, that would imply nearly twice the resolution in the laserdiscs. Looking at those screencaps, that is definitely not the case.

Sorry to confuse everyone.

-G

Mielr said:

EDIT: I didn't know what you meant so I did a bit of digging. The 'weave' method is generally considered to be superior in most cases, correct?

Will you be altering your stabilization scripts now, or is that totally unrelated?

 Unfortunately, the damage is done, and the best that can be done after such an event is already in the script.

-G

Moth3r said:

every other frame discarded? This would explain the vertical blur that is present. 

 Yep.

Half of the vertical resolution was thrown away. It also accounts for all of the aliasing. Conducted additional experiments last night, and now I'm about 99.99% sure.

-G

GOUT was not “weave” deinterlaced, it was “bob” de-interlaced.

still gathering additional evidence, but I’m about 95% sure.

-G

Moth3r said:

Can you post the script? (I'm sure g-force won't mind...)

I wouldn't mind if he wants to PM it to you.

-G

Wow, just wow. You guys are too funny. I love doubleofive's forced subtitles for Chewie and R2 idea, but I love TheBoost's cockney idea just as much. Maybe these aren't mutually exclusive. The accent could be so thick and full of cockney slang that it needs subtitles just to understand it.

-G

g-force said:  

zee944 said:

2. There's some halos around the edges; try YAHR() to reduce them. If you find YAHR's strength a bit too high (losing details), try increase sharpening first and stop right before the halos appear again. If you don't like to result, you may want to use supersampling. Anyway, you'll see what YAHR() does and if it improves your video or not. Probably yes.

Already tried most of what you have proposed. The problem with YAHR (and to some extent even better de-haloers) is that they kill corners and small lines, and leave the video with a soapy, artificial look. One of Didee's poorest products IMHO. I'd be interested in looking at some specific examples if you want to actually try them on the GOUT and post what configuration works for you.

Well, now I've tried EVERYTHING you suggested. Yahr by itself rounds all corners. If you run it on the opening scroll, it becomes almost unreadable. It also kills a lot of stars. Anything that ruins the image this much is not useable.

Tried (apprehensively) with pre-sharpening. Pre-sharpening is usually a bad idea, better to post contra sharpen. Tried both, but neither improved the result enough.

There was one method using YAHR that had some promise.

source = last
med    = MT_luts(source,source,mode="med",pixels="0 1 0 0 0 -1",expr="y")
Median1(source,med,source.YAHR(),chroma="copy first")

Killed the halos, with much less artifacting. BUT, the crawl was still too aweful. If it kills the crawl, it'll kill something else that I care about too. No go.

Tried using supersampling. Doubled the image size and it still destroyed some detail, but it also didn't remove halos at all.

-G

zee944 said:

In the last one day, I've read through this topic - all the 29 pages, yes. First of all, it's very nice to see someone who knows what he does, and can go near the limits. Thank you g-force for your commitment and hard work - and everyone else's work you based your script on.

That's quite some feat. Thanks for the kind words. 

zee944 said:

1. Levels() causes clipping in the video, burning out some details in the brights spots (perhaps even in the not-so-bright spots).

 Levels, if used incorrectly to increase brightness, will cause clipping of the whites. I am not using levels in this way.

zee944 said:

2. There's some halos around the edges; try YAHR() to reduce them. If you find YAHR's strength a bit too high (losing details), try increase sharpening first and stop right before the halos appear again. If you don't like to result, you may want to use supersampling. Anyway, you'll see what YAHR() does and if it improves your video or not. Probably yes.

Already tried most of what you have proposed. The problem with YAHR (and to some extent even better de-haloers) is that they kill corners and small lines, and leave the video with a soapy, artificial look. One of Didee's poorest products IMHO. I'd be interested in looking at some specific examples if you want to actually try them on the GOUT and post what configuration works for you.

zee944 said:

3. From the screencaps I have seen there seem to be a very slight chroma problem. 

Could you be more specific about what chroma problems you see? 

zee944 said:

If it's a smearing similar to color bleeding

No, it's not. See Mielr's explanation above. 

zee944 said:

If it's like fieldblending (that means the trails are coming from the neighbour frames - the smearing led to four-eyed startroopers) 

No, it's not even close to that either. And that is not what led to the four-eyed "STARtroopers".  ;)

The smearing is from a form of noise reduction that is VERY similar to a weighted TemporalSoften(), except using only previous frames, and not having much if any scene change thresholding. This will NEVER (yes, you can quote me on this) be removed from an automatic filtering technique. The major problem is that the filter applied was extremely non-linear. It affected dark scenes more, and can have affects from between 0 and 5+ frames previous. Yes, you read that correcly, I have seen frames that have remnants from at least 5 of the previous frames.

-G

dark_jedi said:

I have an INTEL Quad Core @ 3.0 GHZ,8 GB's RAM(I know WinXP Pro only see's 4 but 4 more are there LOL),Dual Nvidia GeForce 9800 GTX+ Video cards running in SLI mode,Dual 500 GB hard drives,and now an external 1 TB hard drive. 

Beautiful!

-G

dark_jedi said:

But I am not going to make another lossless avi file from the new script until you know for sure it is "Final" LOL. 

Like that's gonna happen... ;)

No promises, but I really do think I'm done for a while though.

-G

ESB fixes incorporated into the script (see page 1)

-G

Nice critique neebis, It's much appreciated.

-G

Mielr said:

g-force, just out of curiosity, I wonder if you see the same thing I'm seeing in the magic tree scene in ESB - which is some "shakiness" in Luke's close-up immediately before and after Vader's helmet exploding (especially after).

I wish I could tell you. I have wiped ESB from my hard drive. So this is not on the original GOUT? Would you be able to check it with the new version of the script?

-G

*INFILE           c:\program files\avisynth 2.5\scripts\stabilize.avs
*OUTFILE          C:\Program Files\AviSynth 2.5\Scripts\stabilize 400.m2v
*BITRATE          4841
*MAXBITRATE       8000
*FRAMES           0 174257
*NOSMP
*PROFILE          best
*AUTOGOP          12
*PROGRESSIVE
*AVSMEMORY        100
*AVSRELOAD
*PULLDOWN
*CUSTOMMATRIX
   8   8   8   9  11  13  14  17
   8   8   9  11  13  13  14  17
   8   8  11  12  13  14  17  94
   9  11  13  13  14  17  17  94
  11  11  13  13  14  17  94  94
  13  13  14  16  17  20  94  94
  13  13  14  17  94  94  94  94
  13  14  17  94  94  94  94  94
 
  12  12  13  14  15  16  22  26
  12  13  14  15  16  22  26  32
  13  14  15  16  22  26  32  41
  14  15  16  22  26  32  41  53
  15  16  22  26  32  41  53  94
  16  22  26  32  41  53  70  94
  22  26  32  41  53  70  94  94
  26  32  41  53  94  94  94  94
*LUMGAIN          2

 

nebis, here's the contents of my HCenc .ini file that shows the custom matrix. Notice I also have the LUMGAIN set to 2.

-G

neebis,

Thank you for the constructive criticism. You have inspired me to try out some things I've been kicking around. I have updated my script to 6.00. Turns out that I've nearly eliminated the problems you're showing with the Stormtroopers in smoke and the Rebel pic immediately following it, and it seems to have not hurt the stabilization too much! Please try it out and let me know what you think. (screen caps are nice BTW)

-G

neebis said:

Please do send me send me your HCEnc matrix when you get the chance. Thanks again.

 It's at home, so I'll send it tonight.

neebis said:

I wasn't really sure about all the values to control the level or noise reduction and sharpness. I tried lowering the sigma to 3 then 4, which is what I'm using. Is that the right variable to lessen the degrain effect?

What about the values below:

source.MVDegrain2 thSAD=400

last.MVDegrain2 thSAD=300,thSCD1=150,thSCD2=75

Will these or other values in that section of the script allow me to fine tune the degrain effect? I plugged in different values (mainly higher), but I didn't get much of an effect that I wanted, so in the end I left them as is.

I also see that Contra is used in the sharpening stage. I really couldn't figure out how to tweak that one, so I didn't touch it. Can you give me a hint about which values to change to increase or lessen the sharpening filter?

Changing the settings in the FFT filter won't have as much effect as you are hoping for. The FFT filter is not "really" used, it is for finding the motion vectors, and then as a check against errors. I would suggest NOT chaging that. Changing the values in MVDegrain2 won't do what you want either. They are also used for error prevention.

Okay, as far as the degrain goes, If you want less, change both the lines with MVDegrain2 to the following respectively:

source.MVDegrain1(bw_vec1,fw_vec1,thSAD=400,idx=2)

last.MVDegrain1(bw_vec1,fw_vec1,thSAD=300,thSCD1=150,thSCD2=75,idx=3)

This will use less mo-comped frames for the degraining. Unfortunately, I've found that if you remove less grain, what remains is too temporally stable (probably from the DVD compression) and tends to look less like real random grain, and more like a "veil" of grain that slowly waves. Also, you will not get rid of as much "local motion", meaning that objects that should be stationary will seem to move in relation to the rest of the frame. Try it out, and see what you think. I'm definitely open to making this thing run faster, which reducing the MVDegrain radius would do.

Okay, if you want more or less sharpening, the best way would be to change the function "contra".

this line:

MT_MakeDiff(last,last.RemoveGrain(12,-1).RemoveGrain(12,-1))

for more sharpening would change to:

MT_MakeDiff(last,last.RemoveGrain(12,-1).RemoveGrain(12,-1).RemoveGrain(12,-1))

or less sharpening to:

MT_MakeDiff(last,last.RemoveGrain(12,-1))

If those don't get you where you want, then I would start changing the "sharpen" value in the FFT.

Let me know what works best for you.

-G

Hey neebis,

If you're using HCenc, I have a custom Matrix that works really well. Interested?

I'd be curious to see what you tweek the values to. Always looking for creative input. Also, any chance you could re-post those split-screens now that you have the script working?

Thanks,

-G

Moth3r said:

Mielr said:

Just for clarification- the script used for these discs stabilized the gate-weave and improved the aliasing as well (like in the PWNAGE edition)? The reason I ask is because in addition to everything else, the aliasing seems improved too.

Yes, g-force's script includes an anti-aliasing stage (it uses the newer NNEDI2 filter, which wasn't available when I did the Pwnage Edition).

 

"Minor correction (something I wrote in another thread): upscaled with EEDI2 and NNEDI! My method has some of the benefit of EEDI2's ability to connect more lines than NNEDI, but making sure not to introduce "blur bubbles" (nor blur the picture). And yes, in case anyone is wondering, NNEDI2 makes no difference with this method."

I think DJs DVDs were made with a version that used NNEDI, not NNEDI2. I hadn't incorporated it yet, as I found it had no effect with my method of anti-aliasing, so I didn't bump the revision. Then I decided to change the subtitles, and just threw in NNEDI2 for giggles.

-G