Need help with strong halos, color bleed (EP16, Neon Genesis Evangelion)

[pwnsweet]

UPDATE: Further clarification in Post #5

Hi everyone.

I'll start with the story for context. In May 2016, Sephirotic and I set out to create the best possible version of Episode 16 of Neon Genesis Evangelion using a combination of video editing, digital enhancement and color correction.

Why is this necessary?
Episode 16 is broken! At some point after the original airing the 16mm negative was lost and Gainax were forced to use a lower quality 35mm internegative for all subsequent releases. This has resulted in all released versions being blurry, crushed, haloed, have incorrect colors and many, many other problems.

A glimmer of hope?
In late 2015 Studio Khara released the Archive of Evangelion DVD box set containing Episode 16 as it was originally aired in 1995! This gave audiences the opportunity to see detail in shadows not seen for nearly 20 years! Despite the increased detail afforded by this release it's also riddled with problems, however with significant editing, filtering, color correction it has the potential to look better than all prior releases, including the Blu-ray. This process started in May 2016. A huge amount of work has been done since then to clean up the scene changes, remove ghosting and splicing it with other sources to create the best possible image quality. 1000+ hours of work and an avs script of more than 3000 lines containing over 5000 individual edits has brought us to a point where we are finally ready to move into the next stage of development of filtering and color correcting.

It's at this point that I make my plea for your assistance. The sources we are working with have strong halos and color bleeding. As such, we need someone with highly advanced Avisynth knowledge to create a specialised script to clean up those edges and remove the bleeding and other artefacts. Before being provided with the source, you will be provided with a sample and be asked to demonstrate your skill.

Nothing like this has ever been done before. It is a gross understatement to say that it has been extremely difficult. My claim of over 1000 hours of work is a conservative figure. Given this, you may be asking yourself whether it's worth it. The answer is an unequivocal 'Yes!'. This is because it is highly unlikely that we will ever get a better version of this episode given that the original negative has been lost to time. Neon Genesis Evangelion has had a significant impact on Japanese and Western culture alike since its release in 1995. The way we see it, by restoring this episode we are preserving an important part of anime history.

Want to help? You can! Here's how
If you would like to be a part of this project and think you possess the necessary skills, we'd like to hear from you. Please download the sample below and submit a link of your completed work to my email pwnsweet@gmail.com. After Sephirotic and myself have reviewed your work, we will contact you.

Thank you!

---> Get the sample here! <--- (https://mega.nz/#!gI4QmAwT!TWhtf_NugpriQ9Oq4Nc6pNsKEcx6tBxIDDPydOOOha4)


http://i.imgur.com/pD4baVQ.png

[pwnsweet]

To clarify:

We have two sources. We want to use a mixture of both sources to create the full episode. Both sources have different (and wrong) colors and Sephirotic has worked tirelessly to fix the colors and make them match. One of our sources has halos, color bleed etc, the other does not. We now need someone to write a script that can bring both sources to the same visual quality. At the very least, this will involve fixing the color bleed and removing the halos from one of the sources. It will also involve sharpening one source and blurring the other, OR sharpening one source to match the other. Which approach you take will depend on your skillset. Either way, the goal is to make both sources match as closely as possible.

[lansing]

I was refer here from your thread (https://forum.videohelp.com/threads/385176-Need-help-from-someone-with-source-matching-expertise) in videohelp, I'll continue here.

I'm confuse with your sample videos and description, where did the "target" images come from? Are they screenshots from the blu-ray or are they your fixes? And what do you mean by two sources? Twos sources of the same episode or different episode?

[poisondeathray]

:confused:To clarify:

We have two sources. We want to use a mixture of both sources to create the full episode. Both sources have different (and wrong) colors and Sephirotic has worked tirelessly to fix the colors and make them match. One of our sources has halos, color bleed etc, the other does not. We now need someone to write a script that can bring both sources to the same visual quality. At the very least, this will involve fixing the color bleed and removing the halos from one of the sources. It will also involve sharpening one source and blurring the other, OR sharpening one source to match the other. Which approach you take will depend on your skillset. Either way, the goal is to make both sources match as closely as possible.



Are both sources the same duration, or does one have different cuts etc...

Otherwise just use the source without the halos and color bleed .... Why would you "dumb down" the good source? What is the point in "matching" ?

Not an entirely clear description... :confused:

[pwnsweet]

I was refer here from your thread (https://forum.videohelp.com/threads/385176-Need-help-from-someone-with-source-matching-expertise) in videohelp, I'll continue here.

I'm confuse with your sample videos and description, where did the "target" images come from? Are they screenshots from the blu-ray or are they your fixes? And what do you mean by two sources? Twos sources of the same episode or different episode?

Hi, and thanks for your response. To answer your questions:

There are two official 'versions' of Episode 16. One was released on DVD (1,2,3 and 4.avi from the sample below) using the original negative during the transfer, and the other was released on BD (target1 - target6.bmp from the sample below) using a lower quality inter-negative. It's important to note that even though the BD version of episode 16 was released in 1080p, it looks much softer than the other episodes that were also released on BD. It looks this way because the inter-negative was scanned for this release at a SD resolution and then upscaled to 1080p. Every other episode was scanned in HD and looks amazing. This is important as we will soon see. It's also important to note that the original cells for episode 16 will never be re-scanned in HD for a future re-release. What we have now is what we will have forever.

So to recap, we have discussed two sources of the same episode: one DVD, and one Blu-ray, both very different in every way except that they both look soft. Our project seeks to create a new version of the episode using the DVD as the base but with selected scenes replaced by those from the Blu-ray. We want to use the DVD as the base because, believe it or not, it's actually better than the Blu-ray in most cases (but not all).

Why would you "dumb down" the good source?
A good question with a complicated answer. Remember how I said that every other episode in the BD set looks amazing? Well it turns out that some scenes from episode 16 are duplicated in other episodes! target7.bmp from the sample linked above is one such example. With these other episodes looking as good as they do, naturally we also want to make use of them to replace those scenes that look bad in both the DVD and BD version of 16 with the equivalent scenes from the other episodes. However because the other episodes were scanned in HD and look so amazing, it's immediately obvious when one of these 'amazing quality' scenes is spliced in and shown during playback of the episode. To combat this, these 'amazing quality' scenes will need to be 'dumbed down' to bring them more in-line with the adjacent scenes from the BD/DVD sources.

We are now at the stage where all the splicing, syncing and color matching/correction has been done and the final form of the episode has been finalised. Many months of discussion between Sephirotic and myself were necessary to ensure that the careful balance of quality and consistency between scenes was maintained. In some cases a particular shot looked better on a different source but we chose not to use it to ensure consistency within the scene. And so with the episode structure finalised, we now need someone skilled in video filtering to create a specialised Avisynth script (or equivalent) to clean up the edges and remove the bleeding from the DVD to make it more closely match the visual quality of the Blu-ray.

If you'd like to help, you can do so by downloading the sample below (~350mb) and:

1. Making "1.avi" and "2.avi" look like target2, target3, target4, target5 and target6 (by removing halos)
2. Making "3.avi" look like target1 (by removing halos and fixing color bleed)
3. Making "4.avi" look like target7 (by removing halos, fixing color bleed and sharpening 4.avi)

Obviously it won't be possible to make them look exactly the same (especially #3), but the closer they look, the less obvious source changes will be. In the case of #3, it's OK to upscale 4.avi and, if necessary, 'dumb down' target7 to achieve a closer result.

[lansing]

The dvd looks more like a vhs transfer to me. And I don't see the dvd having more detail in darker area than the blu-ray. Even though that problematic episode was an upscale, it was still better than the dvd. Why not just fix the blu-ray version?

[pwnsweet]

Hi lansing,

What you have is just a sample. Those samples are not representative of the entire episode with regard to detail in shadows. As already mentioned, the DVD release typically has more color information, less black crush and generally more detail. Here's an example of how different the two releases are in most cases:

https://i.imgur.com/zOsS0fG.png

[lansing]

So the main point of your project is to replace the frames that have crush black in the BD version with the one from the DVD version?

You can probably achieve that by aligning and stacking the BD frame on top of the DVD frame, and then doing some blending mode and opacity tweaking on the top frame like you would do in photoshop to bring out just the dark area. This way you avoid doing all those color correcting and dehalo stuffs.

You may want to test that out in your photoshop first. And also provide the exact frame from the two sources so people can test.

[pwnsweet]

You can probably achieve that by aligning and stacking the BD frame on top of the DVD frame, and then doing some blending mode and opacity tweaking on the top frame like you would do in photoshop to bring out just the dark area.

Even if that were to end in a better result, that's not how we are doing the episode. As I said, the final form of the episode has been finalized. I do appreciate your suggestion though. It shows that you want to help and I am grateful for that. Going forward though, we've decided to tackle the episode differently.

[sephirotic]

I didn't know about this thread. Pwsweet just told me about it.

Let's see if I can clarify things better:

The two sources have both pros and cons.

The Archival DVD source is based on the digital master from TV Tokyo broadcasts. (yes they already used digital media back in 1995). it is NOT a VHS transfer, it is a digital INTERLACED Telecine from the original FRESH 16mm stock film.

This source has the obvious problem of having several blend artifacts, and bad color calibration from the capture machine. Particularly there is a strong mid to dark green cast from a poorly color calibration of the capture machine. Most of the shots also have an yellow (lack of blue) cast on the highlight. The colors are often inconsistent, shifting from shot to shot. The brighter the shot, the yellower it becomes in the highlight, and greener in the shadows. The darker, the color shifts goes to the opposite spectrum: Blueish and magenta. It was obviously NOT A DIRECTOR choice of coloring as the exact same shot interleaved between two has Delta-E changes of up to 5~10 points (In the 15-235 RGB scale I was working in).

One may think this awful source would be nearly useless, but compensating the green and yellow casts would often show up remarkably well how the episode originally was colored like when compared to the second source:

The second "BD" source is scanned from a relatively low quality 35mm interpositive blown up from the original 16mm scan. It was then captured progressively into digital SD. It was later poorly upscaled to the Blu ray resulting in severe Haloing artifacts. The lighting is also pretty poor. The interpositive was not properly exposed, has awful latitude, crushing blacks, and then the source was obviously "pulled up" to the limit of 80% NTSC where the highlights WERE ALSO crushed in a non-linear amount. So everything that is too bright or too dark has the extreme lighting messed up. However, being a fully progressive source and scanned from a 35mm film, this source has no blending artifacts from a telecine capture and is often sharper than the Archival source.

I have experimented in using the 25% darker and brighter portions of the picture from the archival merging with the "BD". While it does work on Photoshop for a single picutre, it is impossible to do it automatically because the film stock sources are different, which means accurately tracking gate weaving and tilt of two different transfers is impossible.

It is easier to just select the best shots of each version, combine then, color correct manually and dehalo everything, which is precisely what we have been doing.

however, using shots from both sources clearly shot the different level of sharpness and fine "texture" as well as the problem of chrome bleeding from the Archival source.
Personally, I don't mind about the blending so much, I personally want to make an archival-only release later, but the Archival also suffers from some haloing.

Anyway for making the two sources go along well, we need to try and make then look alike both in terms of texture, color and sharpness. The biggest problem is the latter: I personally don't want to "reduce" the sharpness of the BD version for it to "merge" with the archival shots, I'd rather use some sharpening on the archival and mask a little the difference between the two sources with some fake grain similar to the 16mm grain we seen on the rest of the episodes from the BD.

In order to get some decent sharp edges without looking badly overfiltered, we need some pretty good dehaloing process. We have managed to get some pretty good dehalo filters that smartly use contrasharpening Technics to remove secondary halos. However, the filter is not perfect and we found a nice guy that offered to improve such filter. For that, however, he would have needed an updated version of awarpsharp2 with some extra features. I'm no avisynth or C++ expert so unfortunately we are stuck until we get the dehaloing (and sharpening) stage perfect. We have already seen some samples of what a proper dehalo filter can do, but without an updated version of awarpsharp we are stuck.

[ChaosKing]

There is also Vine https://forum.doom9.org/showthread.php?t=173703, a good but very slow dehalo filter

[lansing]

I have experimented in using the 25% darker and brighter portions of the picture from the archival merging with the "BD". While it does work on Photoshop for a single picutre, it is impossible to do it automatically because the film stock sources are different, which means accurately tracking gate weaving and tilt of two different transfers is impossible.
What does this mean? Is that you couldn't align the two frames?


It is easier to just select the best shots of each version, combine then, color correct manually and dehalo everything, which is precisely what we have been doing.

I don't get it, you mean appending the two sources into one clip depending on which one is better?



Anyway for making the two sources go along well, we need to try and make then look alike both in terms of texture, color and sharpness. The biggest problem is the latter: I personally don't want to "reduce" the sharpness of the BD version for it to "merge" with the archival shots, I'd rather use some sharpening on the archival and mask a little the difference between the two sources with some fake grain similar to the 16mm grain we seen on the rest of the episodes from the BD.

There's no way to match a crappy quality DVD to a blu-ray even though the blu-ray was an upscale.


Also to pwnsweet, your title right now is pointless. You should change it to something that's related to fixing evangelion for better searching for the eva fans, because only the hardcore eva fans would be willing to do these kind of stuff.

[jpsdr]

For that, however, he would have needed an updated version of awarpsharp2 with some extra features. I'm no avisynth or C++ expert so unfortunately we are stuck until we get the dehaloing (and sharpening) stage perfect.
I've downloaded the source code and begin to take a look.
A lot of asm inside needed to be put outside, but the source code is "short". I would need in the first place a little time to put the asm outside and try to understand how it works, but after this step is done, i can try to see if i'm able to update it with the features you may need.

[sephirotic]

What does this mean? Is that you couldn't align the two frames?

Google gate weaving.
Even the most advanced scene tracking tool wouldn't be accurate enough for automatically aligning two different sources. It would have to be done manually frame by frame.
At this point I'd rather just redraw every single cell by hand and nakin and scan them again. It would probably be faster too.
I actually even considered doing something similar by vectoring and making a "replica" of the movie on AFter Effects. The main reason I gave up (apart from that being an insane amount of work, was because of the bloody rain scene where the cells are not well defined to be properly replicated.)


I don't get it, you mean appending the two sources into one clip depending on which one is better?

Yes, and that has already been done. I have also manually corrected the color shifts of the entire Archival and "BD" version of episode 16.


There's no way to match a crappy quality DVD to a blu-ray even though the blu-ray was an upscale.

It is pretty damn close. The "Blu ray" version is not really "blu ray" in quality, I forgot to mention that the 35mm interpositive was mastered in SD resolution before being uspcaled for the BD. So we are talking about two originally 480p sources here. Biggest difference is the lower latitude and blending artifacts. Blending artifacts were mostly fixed manually by pwsweet on Photoshop.


I've downloaded the source code and begin to take a look.
A lot of asm inside needed to be put outside, but the source code is "short". I would need in the first place a little time to put the asm outside and try to understand how it works, but after this step is done, i can try to see if i'm able to update it with the features you may need.

That was awesome to hear.
I have some friends that work with C++, I asked them to have a look but they say they don't know anything about Image processing so I'm not sure if they would be any help at all.

[jpsdr]

I have some friends that work with C++, I asked them to have a look but they say they don't know anything about Image processing so I'm not sure if they would be any help at all.
Issue is that 80% of the code is assembler, so if your friend is good with C++, but bad with assembler, he can forget it.
And that also where the problem lies. There is no "pure C" version of the algorithms, all the image process code is only assembler (MMX and SSE2 version). I'll drop the MMX version, meaning the requirement will be at least SSE2.
But, the lack of "pure C" will make it very very difficult to understand... :(. Meany very very difficult any change/features.
I'm not against asm, i'm doing it myself, but i always have a "C" version,

[jpsdr]

Stupid question.
Don't you think, instead of trying to fix the result of the upscale, it could be interesting to see if you can "revert" the upscale, and then after upscale with a better method ?
To see if you can revert the upscale, can't these fabulous filters of mine here (https://forum.doom9.org/showthread.php?t=174846) be of any interest ? ;)

[sephirotic]

Stupid question.
Don't you think, instead of trying to fix the result of the upscale, it could be interesting to see if you can "revert" the upscale, and then after upscale with a better method ?
To see if you can revert the upscale, can't these fabulous filters of mine here (https://forum.doom9.org/showthread.php?t=174846) be of any interest ? ;)
Not a stupid question at all!

Those are a nice collection of filters. I actually tried using the famous Debilinear in different set of assumed original resolutions for the "BD" upscale of episode 16, results were pretty sharp but they were not free of artifacts so I assumed it wasn't a pure bilinear upscale and that the artifacts would do more harm than good.

I could try again with a different mixture of your updated filters.

However, the other source, the Archival DVD also suffers from haloing. It would be difficult to detelecine without many blending artifacts and also revert the rescaling to try and remove the haloing. But I think is worth the shot.

By the way, I have just read your PM and will be answering you shortly, I'm very excited of your willing to help with awarpsharp, I will defintely be willing to test it out, but Pwsweet is the guy that is currently dealing with the dehalo, I'll call him out!

Thank you very much.

[pwnsweet]

jpsdr, I cannot express in words just how grateful I am with your willingness to help! Thank you so much. I'll be in contact with you via PM shortly.

[jpsdr]

Well... You can guess it's not totaly pure free... ;)
One of my interest is seeing your project achieved. So if i can help you achieve your goal, it will also benefit me.

[real.finder]

Issue is that 80% of the code is assembler, so if your friend is good with C++, but bad with assembler, he can forget it.
And that also where the problem lies. There is no "pure C" version of the algorithms, all the image process code is only assembler (MMX and SSE2 version). I'll drop the MMX version, meaning the requirement will be at least SSE2.
But, the lack of "pure C" will make it very very difficult to understand... :(. Meany very very difficult any change/features.
I'm not against asm, i'm doing it myself, but i always have a "C" version,

VapourSynth version added C code sometimes ago

they need depthV and blurV so can work with only horizontal halos or colour bled safely without hurt the vertical details by set depthV=0 or depthCV=0, same for blurV
more info https://forum.doom9.org/showthread.php?p=1797208#post1797208

[jpsdr]

Do you have a link to this VapourSynth version ?

[real.finder]

Do you have a link to this VapourSynth version ?

yes, this https://forum.doom9.org/showthread.php?t=172721

[jpsdr]

As it's described in the text, it seems the avisynth and vapoursynth version may very slighty differs.
On the other hand, i would like to try working from the vapoursynth version (instead of the avisynth version), which have a C code source, and may be easier to update.
Is it good for you ?

Edit :
Finaly, i'll see what i'll do... Maybe i'll try "both"... I'll continue the avisynth version for now.

[jpsdr]

they need depthV and blurV so can work with only horizontal halos or colour bled safely without hurt the vertical details by set depthV=0 or depthCV=0, same for blurV

Even if i'm not at this stage yet, i'm tring to think ahead and see what could be done.

For blur, there is nothing like you describe. But, the blur procedure process doing first an horizontal blur, and after a vertical blur.
What could be added is an int "mode" parameter, with the following for exemple :
0 (default) : Blur H & V.
1 : Blur H only.
2 : Blur V only.
This could be done properly without breaking the compatibility.

Sobel : Use to detect edge... I don't think we should change things.

warp & co : There is, at one point, some horizontal and vertical values computed from the edge information, and these values are multiplied by the depth value. It seems easy to test at this point the effect of having a seperate value for h/v instead of the same.
The issue would be the implementation.
A proper would be to replace depth/depthC by depthH,depthV,depthCH,depthCV, but it will break compatibility.
A less proper way would be to just add depthV,depthCV, with default value of 128.
If default value, behavior is the same, otherwise, depth is considered being horizontal value only.

Is it what you want to have, and if yes, does it seem good/interesting for you ?

[real.finder]

Even if i'm not at this stage yet, i'm tring to think ahead and see what could be done.

For blur, there is nothing like you describe. But, the blur procedure process doing first an horizontal blur, and after a vertical blur.
What could be added is an int "mode" parameter, with the following for exemple :
0 (default) : Blur H & V.
1 : Blur H only.
2 : Blur V only.
This could be done properly without breaking the compatibility.


that ok, but it will be better if there are some way to added blurV (if it can done)


Sobel : Use to detect edge... I don't think we should change things.


I think I will agree


warp & co : There is, at one point, some horizontal and vertical values computed from the edge information, and these values are multiplied by the depth value. It seems easy to test at this point the effect of having a seperate value for h/v instead of the same.
The issue would be the implementation.
A proper would be to replace depth/depthC by depthH,depthV,depthCH,depthCV, but it will break compatibility.
A less proper way would be to just add depthV,depthCV, with default value of 128.
If default value, behavior is the same, otherwise, depth is considered being horizontal value only.

Is it what you want to have, and if yes, does it seem good/interesting for you ?

depthV,depthCV with default value of depth,depthC, just like avs internal blur(), and added them in end of parameters list to not broke existing scripts

[jpsdr]

that ok, but it will be better if there are some way to added blurV (if it can done)

Unfortunately not, because the blur parameter has nothing to do with the amount parameter of the internal avs blur.
The blur parameter is just an integer [0,1] to select one algorithm or anoter, both radius fixed and with simple computation easy to optimise in asm.
The amount parameter of the internal avs blur is a float, designed to work with a more complex and generic blur algorithm.

depthV,depthCV with default value of depth,depthC, just like avs internal blur(), and added them in end of parameters list to not broke existing scripts
I don't think it's possible to make a default value for a parameter the value of another parameter.
depth can have -128 to +127, so puting a default value of +128 for depthV will set that if it's +128, depthV=depth, otherwise, it keeps its value.
The actual code for depthC is not even realy proper, because depthC will be computed directly from depth if equal to 0, meaning that you can't put depthC to 0 if you wanted... :(
A fix would be to do the same, default value to 128, and computed from depth if value is 128.

[real.finder]

Unfortunately not, because the blur parameter has nothing to do with the amount parameter of the internal avs blur.
The blur parameter is just an integer [0,1] to select one algorithm or anoter, both radius fixed and with simple computation easy to optimise in asm.
The amount parameter of the internal avs blur is a float, designed to work with a more complex and generic blur algorithm.


I think that type

int blur = 2

Number of blur passes over edge mask. Less passes increase sharpening effect, but can produce major artifacts with high depth and thresh values.
You can use values higher than 100, but probably won't see any difference. Chroma is processed with (blur+1)/2 passes.
Range: 0 to 100

Note: Unless blur is specifically set, it will default to 3 if type=1.


int type = 0

Type of blur:

0 : radius 6 blur (some kind of 13x13 average).
1 : radius 2 blur (some kind of 5x5 average), requires around 8x more passes than type=0 for the same effect (will be 2.5x slower), but will produce better quality.

[jpsdr]

You're right, i've mixed things up, i was too much focused on some small parts...
So... The following is possible :
Add an int blurV, default value of -1.
If bluV=-1 => blurV=blur, same behavior, otherwise, blur is the number of horizontal passes, and blurV is the number of vertical passes.

[real.finder]

how http://avisynth.nl/index.php/RgTools/RemoveGrain did this

int mode = 2
int modeU = mode
int modeV = modeU

why can't this be done in awarpsharp2?

about if depthC=0 will use depth, I think it's better to throw this behaviour, and make it don't process chroma and just copy it from the source instead of that

[jpsdr]

If you don't set a parameter X, this one will take a default value, this default value will tell the parameter X to take the value of parameter Y...
So, where is the problem ?

In aWarpsharp :
About not processing the chroma but just copy it, there is no such thing actualy in the code.
And it could be done this way only if there is no blur on chroma, but actualy, the blur parameter is for both luma and chroma.
Maybe add blurC/blurCV parameters, to eventualy being able to specify a different blur level for luma and chroma ? I think it could be done easly.

[real.finder]

If you don't set a parameter X, this one will take a default value, this default value will tell the parameter X to take the value of parameter Y...
So, where is the problem ?

no problem, I just think about


Default values
int depth = 16
int depthV = depth


in first place, or like what avs blur() did, but you said that not possible in previous posts

[jpsdr]

Would blurC/blurCV something interesting to add in aWarpshar and aBlur ?

[real.finder]

Would blurC/blurCV something interesting to add in aWarpshar and aBlur ?

why not? any new parameters are welcome :)

maybe you can take another's from https://pastebin.com/09sXzKfa too

[jpsdr]

The only thing more in it it's bic, but it's not realy possible, we'll have to live with what is actualy implemented.
Send me your feedback by PM after you test what i've send to you, this way, i'll be able to read it sooner.

[real.finder]

The only thing more in it it's bic, but it's not realy possible, we'll have to live with what is actualy implemented.
Send me your feedback by PM after you test what i've send to you, this way, i'll be able to read it sooner.

oh, I just note you send something in gmail since 2 hrs, will see it now, thanks

[jpsdr]

BTW, as you may notice, the dll is called "MT" at the end, because i intend in final step to add internal multi-threading.
I intended since the begining to rename the function with "MT" at the end, as i've already done with some of my plugins, to separate them from their original versions.
I didn't do it until now to make things more practical to eventualy test with existing scripts, but from now, the functions will be renamed with "MT" at the end.

[pwnsweet]

Just wanted to chime in here and say thanks for all that hard work you're putting in to this jpsdr and real.finder! I want you both to know that I really appreciate it.

[real.finder]

BTW, as you may notice, the dll is called "MT" at the end, because i intend in final step to add internal multi-threading.
I intended since the begining to rename the function with "MT" at the end, as i've already done with some of my plugins, to separate them from their original versions.
I didn't do it until now to make things more practical to eventualy test with existing scripts, but from now, the functions will be renamed with "MT" at the end.

internal MT are welcome, but will you added both in same dll or separated dlls? or you will not do non MT one?

[jpsdr]

No, i will not do several dll, just the MT one.

[real.finder]

No, i will not do several dll, just the MT one.

then in this case it will be better if you don't rename it

[Katie Boundary]

I find it odd that there is no mention of the Renewal box set that was released in 2003. Seeing as how this was the "version" of the episode from which the Blu-Ray release was upscaled, why not upscale the Renewal release yourself using something like Spline144? That would solve the issue mentioned earlier...

Don't you think, instead of trying to fix the result of the upscale, it could be interesting to see if you can "revert" the upscale, and then after upscale with a better method ?
To see if you can revert the upscale, can't these fabulous filters of mine here (https://forum.doom9.org/showthread.php?t=174846) be of any interest ? ;)

Not a stupid question at all!

Those are a nice collection of filters. I actually tried using the famous Debilinear in different set of assumed original resolutions for the "BD" upscale of episode 16, results were pretty sharp but they were not free of artifacts so I assumed it wasn't a pure bilinear upscale and that the artifacts would do more harm than good.

If you don't have the Renewal box, I can send you the raw VOBs.

[real.finder]

I find it odd that there is no mention of the Renewal box set that was released in 2003. Seeing as how this was the "version" of the episode from which the Blu-Ray release was upscaled, why not upscale the Renewal release yourself using something like Spline144? That would solve the issue mentioned earlier...


even if they don't have that edit (maybe they have it), I don't think it will be any better in this ep, Spline144 is not the best upscale btw

anyway, post some sample of the vob here to check

[pwnsweet]

I find it odd that there is no mention of the Renewal box set that was released in 2003. Seeing as how this was the "version" of the episode from which the Blu-Ray release was upscaled, why not upscale the Renewal release yourself using something like Spline144? That would solve the issue mentioned earlier...


I can answer this one. We have the Renewal release but, like the Blu-ray release, it too has severe halos. The Blu-ray release and Renewal are the same, except the Blu-ray has slightly more detail so it makes sense to use it instead of the Renewal.

But there is another reason we chose to use Blu-ray and that is convenience. Before this project, Sephirotic ran a project to color correct and filter the Blu-ray release with the help of Fussoir, a well-known anime encoder. The script Fussoir wrote for the episode was extraordinary and accounts for primary and secondary halos, color bleed, aliasing etc - all on a per-scene basis. It's truly amazing. The end result is a highly tailored script resulting in a near-perfect image for basically every shot of the episode. Examples of this work are in the sample file linked in the OP (target 1-6). Sadly, contact with Fussoir has been lost and so we cannot ask him to write us another amazing script to filter the Archive which is why I've created this thread.

In light of this and with all the filtering work already done for the Blu-ray it doesn't make sense to use the Renewal.

[Katie Boundary]

The script Fussoir wrote for the episode was extraordinary and accounts for primary and secondary halos, color bleed, aliasing etc - all on a per-scene basis. It's truly amazing. The end result is a highly tailored script resulting in a near-perfect image for basically every shot of the episode.

So then why do you need Archive...?

[pwnsweet]

So then why do you need Archive...?

The comment regarding 'near-perfect image' refers to halos and color bleed being nearly completely eliminated. However the BD/Renewal still suffers greatly from crushed blacks and other issues. That said, some scenes from the BD are still better than the Archive, but only some scenes. For the majority of scenes the Archive is better. So we need a good script to filter the Archive so that it matches the quality of Fussoir's filtered BD as closely as possible so that the final spliced result is as consistent in overall quality as possible.

[jpsdr]

So, what is the actual status of this project ?

[pwnsweet]

So, what is the actual status of this project ?

Basically finished. Just waiting on Sephi to apply the finishing touches and then we're done.

[manofkuwait]

did you ever get those finishing touches? :D

[gugglu]

hi , maybe this can help.

MCaWarpSharp3: by zee944.
This is a bit of all but it can be included in the line enhancer category.
It is perfect for those footages that look like seen through a lens (i.e. blurry, chroma aberrations, halo, etc) but it can also work for only blurry footage.
It is a bit extreme so use with caution.
It's also slow so my modification was mainly for speed improvements although I also reorganized everything a bit.

It's from Dogway slight mods :D
DL Link: http://www.mediafire.com/file/ty30nob26ha0mil/MCaWarpSharp3.zip

[jpsdr]

Yes, what the status of the project ?