@Peter1234
I already asked incredible about this in a previous post...

Change this line:
to this:

Here's my take on the 704x480... There is absolutely no point in changing this back to 720x480, because all that is going to do is add 8 pixels of black border on both the left and right edge of the screen. Since your TV set in almost all circumstances cannot see those extreme left and right edges, it's best to simply leave them out since 704x480 is a perfectly accepted dvd compliant resolution. This way you'll have more space for encoding your movie, because keeping an unnecessary black border in an encode will take up more space than none.

Does this make sense? Am I correct in this assumption?

@J-Wo
Yes, it sounds like you're making sense! :) I've never thought about it in that way before. Good point :wink: .

BTW I'm getting pretty ugly ghosting with this new version of ADS... :cry: Very noticable on scene changes!

J-Wo,
I agree that there is no point in changing 704x480 back to 720x480 after it has been reduced, but it would be better to do all of the processing at 720x480 so that none of the image information is lost. If there was no advantage to using 720x480 then all of the DVDs would be 704x480. When the image is displayed on a 16:9 HDTV those extra pixels are used.

@audioslave,
Thanks, I missed that.

@Audioslave:

Wow, I'm glad I was actually able to assist someone!!! :D

@Incredible:

I would like to see some modification to the way you handle the cropping feature. My HD set can handle an overscan of 3 (24 pixels) on the left and right edge, but it's slightly too much on the top and bottom edges when viewing anamorphic 16:9 video. Optimally I can cut off 20 pixels from the top and bottom, so all my encodes lately have used the line crop(24,20,-24,-20). However I can't be so precise with ADS, because while it lets me enter 20 pixels for top and bottom (To, Bo), I'm only allowed to enter overscan for left/right. This means if I chose overscan 3, I'll essentially get crop(24,24,-24,-24) because it will override my To,Bo settings. As a compromise I enter overscan 2, giving me crop(16,20,-16,-20). I don't know if an extra 8 pixels removed on left/right will make that much of a difference in my encodes, but since you already allow the user to specify To,Bo why not Lo,Ro for Left/Right borders instead of Ovr for Overscan?

@ inredible (or others!):

okay I have a problem using ADS with DVD-RB/RB-Opt. Some lines are too long for DVD-RB to be entered in its filter editor, because by god they have a stupid character limit! This splits up long lines with a carraige return, destroying the whole function. When I load the thing up in RB-Opt, I could fix each line but because it doesn't have a full text editor I have to add/delete lines one at a time. PAIN! Can someone explain how I save the entire ADS function into a plugin that I place in my Avisynth 2.5/plugins directory? That way when I call ADS all the code is there and I don't have to put it in my script. Thanks!

@J-Wo
I am using this:
In other words: Save the ADS script to an *.avs or *.avsi file and import it in your script :wink: .

Thanks to both of you, it worked like a charm. :D

J-Wo,
Another option. If you copy function into a text file using notepad and save as ADS.avsi and then put that file in the plugins folder inside the AviSynth program folder, you will be able to call the function using ADS(0,0,0,false,6,8 ), or what ever options you want inside parentheses, without needing the Import statement.

@J-Wo
No, it's not a typo. I saved the ADS script as "ADS-2004-09-22.avsi" :) .

J-Wo,
audioslave's suggestion will work with .avs or .avsi ending.

@Peter1234
Yep! What's the difference between the two extensions anyway?

@incredible
Any news on the ghosting?

Well in my tests I didn recognise any ghosting 8O
But that depends for shure on the source!
So do tweak the temporalsoften(...) line in the script as you like it :wink:

According to 720px encodings.

As explained before ...

Example in PAL:

720 width gots an pixelAspect ratio of 128/117. Thats the value your PAL video will be stretched by your TV.

So 720x(128/117) results in 788 px! But a PAL TV only shows 768 px (seen digitally) so 20 px are OFF, EVEN if a TV would have NO overscan ;-)

Same logic in NTSC, just another pixAspectratio has to be used for calculating.

@ J-Wo

Just use your classic resizing (keep it mod 8! ;-) but you should even resize in FitCD or MS using resize MOD16! generally ) and use in ads() as first 3 values 0,0,0 where you add the borders afterwards.



I updated the function (leaving still the last one in here present so in case of ... ;-) )
I had a nice exchange with dideé at forum.gleitz.de where he recommended his Double Sobel Mask which gives even better detail detection by beeing still resistant against noise.

So the syntax is now:

ADS(int "Top", int "Bottom", int "Overscan", bool "Letterbox", int "Sharpen", int "Threshold", bool "HQmask", bool "Show")

The HQmask and Show parameter have been added.
HQmask=true means the usage of didées double sobel mask (slower but even more detailed!!!)
Show = true means, that the peview of the masked area is shown.

Example:
ADS(72,72,2,Letterbox=false,Sharpen=3,Threshold=4, HQmask=true,show=false)


"Masktools" Plugin is needed for this Function.
http://jourdan.madism.org/~manao/MaskTools-v1.5.1.zip

Inc: regarding values for Sharpen and Threshold... I've noticed that Threshold values above 2 smooth out too much detail on my DVDs. If I turn HQmask on should I then raise my Threshold? I also don't seem to notice as much effect of the Sharpen parameter than the Threshold one. But I have to say so far I like what I've been seeing.

incredible,
Thanks for generating all of these functions. I did some more tests with your previous version using ADS(sh=6,th=12) on DV files with convert60ito24p() and I think it does improve the image quality. I also tried your newest version using ADS(0,0,0, Letterbox=false, Sharpen=6, Threshold=12, HQmask=true, show=false) on the DV files but I did not think the results were as good as with the previous version. No quantitative data to support that, just my impression. I assume that this only works on progressive sources, so if I want to use it on an interlaced source I should separate the fields and then use ADS on each field. Is that correct?

Well the diff. between the previous and this version is that you got the option to choose dideés doble sobel mask ("HQmask") beside the normal mask.

You also can rise the threshold of the HQ mask if you rise the first value in Levels() of the function (ill integrate that as a parameter).


According to interlaced processing you should use


bob()
ADS(.......)
Separatefields().selectevery(4,0,3).weave()

I do set 4,0,3 as if using 4,1,2 the final fieldorder gets changed in case of bob(). Bob() is the choice as its stupid and not smart= best condition for handling the noise.

A perfect solution would be:

Sourceimport()
bob() # stupid bob
Denoisers(.......)
Separatefields().selectevery(4,0,3).weave()
Kernelbob() # or other smart bobbers
Resizers()
Separatefields().selectevery(4,0,3).weave() # or 4,1,2

BUT as ADS() does both denoise & resize I will integrate an interlaced option so the optimal bobbing will be intergated in the script.

you also should everytime check your fieldorder first before bobbing.

8O :rotf: why " stupid bob " inc ? :lol:

@inc: Can I propose you to open a new thread for your function? this way it gets more transparent to people who want to try out your script. You could post your script, together with a short explanation on how it's supposed to be used/for what target and so on... also updates could be posted in that new thread more easily! just a tip! :wink:
good work, I'll try the function out very soon!

A stupid bob just "seperates" the fields by keeping the fields position which seperatefields() doesnt and resizes using the bicubic std. interpolation.

A smart bobber does work adaptive to combings means does handle real combed areas different using interpolation between fields for instance = the XXXbob output will be more manipulated (said in easy words). And there are many diff bobbers containing each their diff. advantages. TmcKernelBOB(), TmcBOB(), KernelBob, MatchBob(), Tdeint(), DgBob() and so on, some of them are provided as dll's and some as functions based on tomsMoComb() for instance.

@ Jell (and Jorel)

I do see that its mega senseful to split this thread and so Ill do it :D

Yesterday I did some tests using diff. Temporaldenoisers and sharpeners, I do think Ill release the next beta this evening.

This script could be used for both, DVD inputs and captures, as everything depends on a right setting of the thresholds!

So the best way is to watch HOW much has the threshold to be rised till the noise is off, and that also can be figured out by watching the regular preview. This will keep as most detail as possible.
On DVDs you very often get more "spray" noise than like in captures where a "grainy" noise is oftener to be found.
Now as grainy noise itself is bigger, so you have to rise the threshold. And this wont make that much harm as captures do already have much less details then DVDs.

The only problem on captures is, that also the masked edges do contain very visible "grain" in the unfiltered state. So if sharpen that masked edge area of captures, mosquitos will result as remarked by Didée at forum.gleitz.de. So I've to find a spatial filtering solution in case of edges from captures! But already of no sharpen in case of captures is applied, the function does wonders when using the right threshold.

As said the threshold setting now only works on the normal mask mode.
If you want to enter the threshold of Dideés "HQmask", then do rise the threshold in the Levels(x,1.0,255,0,255) command where x could be rised or also or only the gamma ( the 1.0 value in levels() ).
Do use show=true and youll see how the threshold will affect the mask.


Yesterday I took a sliced part from "a new hope" of the starwars box set. The noise was off, details where even more enhanced and the compressibilty was ended up in ca. 68 MB, where undot().deen() gaves me 65 MB and so that little amount of filesize rise Ill take into account for myself as it stands in a relative GAIN of the quality rise. So u can assume how tiny will be the rise in Q of CCE.

very cool ink. :)

Im also thinking about to build i a special motion adaptive option (on/off parameter) if desired.
Means The masked plain areas will get a very spatial softening on fast scenes. while the details will be "cached", means you wont have that MA phenomenom when a girl stands on the street and a big truck passes by in the backround which actually in the present MA would smooth the whole frame incl. the static sharp girl in front.
For shure when "caching" these Details less compression will be the result compared to the actual MA .... BUT the risk of such phenomenoms like mentioned above is much more reduced.


So im also thinking about doing a modular job, means not only one function but an avsi file which contains several functions which could be single! called and which also will be called within ads().

Second I have to get into that building the convolution kernel of degemask() where the mask is not called from presets like "roberts" but from 3x3 Kernels using values like shown above in didées dedgemask call within ads().

Also an optimzed script could be done via the integration of masktools, means if very complex scenes do occur (detection via YdifferenceToNext(edgemask) ) an "nf" adaptive "cpu2="ooooxx", fmin(quant=nf ,32) would be performed.
hmmmm

May I ask what the difference is between ADS, and Fluffbutt's script (can't remember the name right now) that he made maybe half a year ago?
I know it used MaskTools too... :roll:

It was HybridFupp from Fupp. I just looked into the source but its approach in filtering is different. It shurely got his on pros and contras like mine, but my resizing is based on One-half-one.

Masktools ist just like the name says a plugin which makes avisynth capable using enhanced masks. But everything you do IN the masked areas is script specific. Also its an own story HOW do you generate masks.
Didée also gots nice functions which do base on masktools like Limitedsharpen(), Restore24(), ipp().

;-)

Hi All,

This is my 1st post and I must say that I am hooked. You guys are really awesome. Kudos to everyone for making this possible.
Meanwhile, I have been following this thread and had perform some test and interestingly HybridFupp crops up.
Here is a test I did (side by side .... aka stacking) comparing ADS and HybridFupp. For ADS I am using the latest function posted.

Here is the screenshots

http://www.digitalfaq.com/archives/error.gif

http://www.digitalfaq.com/archives/error.gif

I did a 210 sec test individually .... i.e. each test for ADS and HybridFupp

Processing time:-
ADS - 10 mins 47 sec
HybridFupp - 9 mins 50 sec

Size:

ADS - 53MB
HybridFupp - 64MB

BTW, I incorporate kwag's Linear Motion Adaptive Filtering in the avs script that uses HybridFupp.

So inc ... perhaps we can look forward to improvement to ADS processing time and sharpness i.e. reduce blurring.

Sorry should my posting is irrelevant.

Cheers!
Scoprio

scorpio,
Great first post. But, I think you are comparing apples and oranges. Just to see what would happen, I replaced ADS(sh=6,th=12) with HybridFupp() in my script. The results were better with ADS when using my noisy DV source tape. Sorry, I have no way to post nice pictures like you did. I think each filter has its place and needs to have the correct settings to get good results.

Hi Scorpio,

it seems that you did not set the mask threshold right. As you can see above in my samples I never got that blurry pics when setting the threshold right.

EVERYTIME do first make a preview using show=true, there you can see jow much the threshold has to be rised till the noise is off and beside this to still keep details.

Several times in the forum I have seen threads on new functions, so the cases of hybridfupp and now ADS. Never I have seen reach a definitive conclusion nor that they are gotten up to optimal scripts section. It could be possible this time?
Or it means that hybridfupp and ADS does not obtain the quality of the Kwag's optimal MA script.

Hi Pete / Inc .... Thanks mate.

Pete are quite right to point that "each filter has its place and needs to have the correct settings to get good results" and that is the absolute truth.

Thanks for the suggestions and I have done a couple more test and data collected as follows:-

As u can see ADS is very promising. I am currently running the script with DVD-RB and RB-Opt.

Samples Shot with Sharpen(6) Threshold (12) - HybridFupp

http://www.digitalfaq.com/archives/error.gif

http://www.digitalfaq.com/archives/error.gif

I am totally blown away with ADS output and the significant time saving.

Cheers Guys!
Scorpio

Hi,

The only problem I see, is that the actor has been shaved by ADS...

Can you post the same picture from the VOB?

Thanks,
Fabrice

scorpio,
Sharpen(6) Threshold (12) are deffinitely not a good choice for high quailty video. As you can see the detail is removed because that is a setting to remove a lot of noise. I have not investigated settings for improving detail on high quality video, but they would be more like Sharpen(8)Threshold (4). The large Threshold value is going to smooth out detail in order to remove noise. One of the advantages of this function is that it can smooth the solid areas without bluring the edges.

great observations fabrice! :wink:
i see and think the same, deserve a little more sharpness(maybe sharpness), a vob picture for reference and link(or thumbnail) to see the result in 720x??? (full size).will be the best way for comparisons!

scorpio
if need space to host big pictures use: http://www.imageshack.com
:)

you can do something like this:
http://www.digitalfaq.com/archives/i...2004/09/32.jpg
click on thumbnail or in the link: http://img61.exs.cx/img61/5271/jorelcomputer.jpg

sorry..i forgot to post:

:arrow: ADS quality is fantastic, IMHO only need little more sharpness.
question of personal taste but if the picture could be posted in full size,
the sharpness will encrease and if it happen, is perfect! :wink:

Guys thanks for all the input .... I am taken ill since yesterday so here are a couple of screen shots including the source vob.

Source Vob
Sharpen(8) - Threshold(4)
Sharpen(8) - Threshold(3)
Sharpen(8) - Threshold(2)
Sharpen(6) - Threshold(4)
Sharpen(6) - Threshold(3)
Sharpen(6) - Threshold(2)
Sharpen(4) - Threshold(2)

Just click OK when prompt for password.

IMO a lower threshold produce an output comparable to the source. Honestly I am very happy with ADS output .... perhaps I an just a bit lazy to play around with other tunables parameters.

Cheers!
Scorpio

scorpio,
Great. Thanks for posting all of the test photos.

Yesterday I was digging around in my stock of functions and script and I only found a very old version of HybridFupp. So I downloaded the latest release of hubfupp and had a look at it.

On clean DVD Sources it does a very nice job! (as known)
But on captures you have to rise a lot the values of the choosen temporal denoiser. This caused a lot of temporal artifacts. Thats NOT the fault of HybFupp but of the regular side effect when hard temporal filtering a lot of noise. And thats what ads() tries to avoid by reducing the height by 2 before it gets denoised, where the details will be overlayed again afterwards using the maskedmerge.

So as said in the beginning of the thread.

1. - Do set the source path in your avs and do set in ads() show=true
2. - Look at the mask. Do lower the threshold till just a bit of noise is included in the mask (lowering the threshold is don via threshold=x or in case of HQmask the first value of levels(x,.....) in the function. This gives you the best efficience of keeping details by still handling the noise.
3. - do set show=false and do your preview again.

Related to captures:
BE CAREFUL with the adding of sharpeness as captures do also got a lot of noise on edges. That means if just sharpen those edges, mosquitos will be the result.
And thats why still no new version of ads() was relesed this Weekend: I was searching for a best possible mask method based on an optimal convolution kernel.

Developement is continuing .... ;-)

you're not alone my friend! :wink:

of course i only saw the pictures but let me ask to ink:

PLEASE.... :bowdown: show me the script for dvd sources! :D
i want to use ADS now in my tests!

Hi Inc .... I attached here 2 shots with 2 different TH value, sharpen fixed to 8, HQMask=false and show=true.

Appreciate if you could enlighten me which shots has lower noise in the mask.

TH2
TH12

Jorel ... I think Inc is referring to HybridFupp (I may be wrong though).

I did another test with the same source ... S(8), TH(4) ... I get a file size of 160,892,928 processing time 10mins 14secs

I also did another test with the same setting as above (S(8), TH(4)) and now I include kwag's LMAF in the script .... this time I get a file size of 136,329,912 processing time 14mins 28secs

Cheers!
Scorpio

:douh:
oh yes :oops: ...thanks! :wink: your fault scorpio. (kiddin)
i read faster after see your great pictures posted! :lol:

hummm i see...then, redoing a question:

ink, show me the ADS script for clean DVDs sources please!
:bowdown:

@ Scorpio

First:

As you can see in the first mask the "white" areas do show which will be kept as detail (and if set which will be sharpened)! All other areas will be denoised/smoothed --- so the mask with higher threshold will result in less "white" areas --- means less details kept (or in other words more low details will be smoothed)


Second:

As you do sharpen the source's details for shure the compression will be less. So its all upon a compromise ;-)

What do you mean by LMAF ??? Did you include the MA routine??
What means "include"??? Doing a separate test ONLY using the MA?
BECAUSE: DO NOT MIX diff. functions!!!!!!!!!!!
The same thing if people try to enter the "optimalscript" from DialHot, changing parameters/allocations etc. and then the problems which do result are messed up outputs!


@ Jorel

On clean DVD sources??
Well EVERY "clean" DVD source gots diff. amount of noise, so do the testing using diff. Thresholds and do preview the avs in Vdub.
Start from threshold=3 and do rise it till the noise is off AND all needed details are kept. :wink:

The real link:

http://www.imageshack.us/