Restoring the hellhole that is the Evangelion Episode 16

[rrats]

Evangelion is an anime from 1996 and over the years it has received 2 or 3 telecines, with one of them being the one made for broadcasting back in 96 and the other one for the streaming services and bluray release in 2015. I believe there was also one done in 2003, anyway,

between 1996 and 2003 the original 16mm negative for the episode 16 had been lost, possibly stolen, so for the bluray restoration a 35mm print had been used, which looked terrible since the black levels were really messed up and stuff (that might have also been a fault of theirs). The best looking version rn in terms of shadow detail is from "Archives of Evangelion", and that one also looks like junk.

Now is there any way to fix some of many problems present in this video, such as the odd frame blending, uneven telecining and haloing?


[Attachment 64824 - Click to enlarge]


[Attachment 64825 - Click to enlarge]

[pcspeak]

This doesn't solve your problem.
Contains a lot of great info about Evangelion.
It may help, or just lead you further down the rabbit hole.
http://www.users.on.net/~attema/Pen3/faq/#redone

It is somewhat out of date.

[SaurusX]

This was discussed at doom9 a good while back. I don't think the project discussed there was ever completed. But it's interesting to note that many other people have noticed the problems.

https://forum.doom9.org/showthread.php?p=1819882

[october262]

Evangelion is an anime from 1996 and over the years it has received 2 or 3 telecines, with one of them being the one made for broadcasting back in 96 and the other one for the streaming services and bluray release in 2015. I believe there was also one done in 2003, anyway,

between 1996 and 2003 the original 16mm negative for the episode 16 had been lost, possibly stolen, so for the bluray restoration a 35mm print had been used, which looked terrible since the black levels were really messed up and stuff (that might have also been a fault of theirs). The best looking version rn in terms of shadow detail is from "Archives of Evangelion", and that one also looks like junk.

Now is there any way to fix some of many problems present in this video, such as the odd frame blending, uneven telecining and haloing?


[Attachment 64824 - Click to enlarge]


[Attachment 64825 - Click to enlarge]

found this - https://www.reddit.com/r/evangelion/comments/r1ligl/an_attempt_to_restore_episode_16/

[rrats]

I tried mixing both the downscaled bluray (too much contrast/black crushing done) and archives dvd, with both of them being re colour corrected.



(this ones from original telecine)

[Selur]

Aside from the color correction (I usually do this in the last step and it's mostly is a matter for preference), do both DVD and Blu-ray have the same ghosting issues ?
In the clip.MPG there is basically no frame without ghosting on horizonal scrolls,...

jsut for fun, Dehalo_alpha + VSGAN(2x_SHARP_ANIME_v1):

[lollo]

Excellent result! What parameters did you use for dehalo_alpha?

[Selur]

I used the default settings, couldn't really find that changing from the defaults did a noticable difference.
VSGAN model I used can be found over at https://upscale.wiki/wiki/Model_Database

Cu Selur

[jagabo]

Try something like:

edgemask = mt_edge(mode="prewitt", thy1=100, thy2=200).mt_expand().mt_expand().Blur(1.4).GreyScale()
Overlay(last, Dehalo_alpha(rx=4.0, ry=3.0, DarkStr=1.0, BrightStr=1.0), mask=edgemask)

[Selur]

(whole cleaning up the source doesn't seem much of an issue to me, since everything I saw could be tackled through different filters,...)
@jagabo: got an idea how to get rid of the ghosting in the horizontal scrolling scenes ?

[poisondeathray]

For the horizontal pan ghosting, if you use only the 1 source, you redo parts of the animation using a clean plate

For example, the gun pan shot has every 5th field almost clean (well, much cleaner than the others), so you either motion track + composite (e.g. after effects), or interpolate (probably artifacts, not as good choice because this is a pan over static contents)

https://www.reddit.com/r/evangelion/comments/btoqje/digital_enhancement_of_episode_16/

That one used the crossfade, so you can see the previous scene in the right. I would have cleaned it up a bit more in photoshop, because there is still minor color bleed and ghosting

[rrats]

do both DVD and Blu-ray have the same ghosting issues

No, but due to the extreme black compression some scenes can be deemed unusable. Such as this one https://www.twitter.com/LettererBrendo/status/1145870055936286720/photo/4 (photo 4). Tweet illustrates the difference between two versions pretty well. I tried messing with premiere pro and Photoshop and getting any detail out of Asuka in that scene is almost impossible.

[lollo]

I used the default settings, couldn't really find that changing from the defaults did a noticable difference.

Thanks!

Try something like:

edgemask = mt_edge(mode="prewitt", thy1=100, thy2=200).mt_expand().mt_expand().Blur(1.4).GreyScale()
Overlay(last, Dehalo_alpha(rx=4.0, ry=3.0, DarkStr=1.0, BrightStr=1.0), mask=edgemask)

Thanks! A mask is always a good idea, in particular with dehalo_alpha filtering, but in this specific case, because the nature of the video I do not see details removal or artifacts in Selur's samples. I will experiment on the whole video by myself, it can be a good exercise also for real life videos.

[Selur]

The upper images do not seem unusable to me,... got a small sample of that scene?

[poisondeathray]

but in this specific case, because the nature of the video I do not see details removal or artifacts in Selur's samples.

There is some detail removal - Look at the smoothed grass textures on the hill, building textures

But there are different "camps" as to what "anime" should look like. Some would choose to preserve more the original anime look and artstyle.

Another model that could be used with pre dehaloing is "2x_LD-Anime_Skr_v1.0", it actually dehalos too but will fix other remaining issues like color bleed from the dot crawl and it tends not to reduce line width as much (awarpsharp look) . It depends on the look you are going for

[lollo]

There is some detail removal - Look at the smoothed grass textures on the hill, building textures

Correct.

But there are different "camps" as to what "anime" should look like. Some would choose to preserve more the original anime look and artstyle.

Indeed. However, my final goal is to work on analog captures (not animation) having halos because the original masters used in broadcast, so I cannot introduce too much details loss.

Can you reccomend a model with a pre-dehaloing of not animation material?

Thanks!

[rrats]

The upper images do not seem unusable to me,... got a small sample of that scene?

If you are referring to my message look at the Twitter post I've linked.

[Selur]

Am I missing something in the Twitter post? I see VLC screenshots, but no video sample.

[poisondeathray]

However, my final goal is to work on analog captures (not animation) having halos because the original masters used in broadcast, so I cannot introduce too much details loss.

Can you reccomend a model with a pre-dehaloing of not animation material?

Not really , it depends on the source , but nothing works that well on severe live action halos. If there are mild halos, you might be able to improve it

The halos in analog captures tend to be much wider and severe than typical anime halos. Also the smooth cell BG's and shading are more forgiving in anime to "dehaloing" filters and models , whereas a live action source will usually be butched from the dehaloing (conventional means or otherwise)

[Selur]

Another model that could be used with pre dehaloing is "2x_LD-Anime_Skr_v1.0", it actually dehalos too but will fix other remaining issues like color bleed from the dot crawl and it tends not to reduce line width as much (awarpsharp look) . It depends on the look you are going for

Just tried, "2x_LD-Anime_Skr_v1.0" and it doesn't seem to touch the halos for me at all,... :/
(But DeHaloAlpha+Masking works fine and sharpening and cleaning should be possible with tons of filters.)

whereas a live action source will usually be butched from the dehaloing (conventional means or otherwise)

agree dehalo on live action content can be a real pain and usually requires tons of tweaking of masks to not loose tons of details.

[poisondeathray]

Another model that could be used with pre dehaloing is "2x_LD-Anime_Skr_v1.0", it actually dehalos too but will fix other remaining issues like color bleed from the dot crawl and it tends not to reduce line width as much (awarpsharp look) . It depends on the look you are going for

Just tried, "2x_LD-Anime_Skr_v1.0" and it doesn't seem to touch the halos for me at all,... :/

It does dehalo small halos, but try it with pre dehaloing ; ie. just replace 2x_SHARP_ANIME_v1 in your previous script if you wanted less texture erosion, less line thinning, more color bleed fixing . You can mix them too - either afterwards by merging the results, or by interplolating between the models to generate a new model

[Selur]

Ah okay, small halos. Yup, might work as sharpener.

[Selur]

It all depends on what you want , what you consider actual details what noise:
DeHalo + Toon:

DPIR(10)+DeHalo+Toon:

DPIR(20)+DeHalo+Toon:

DeHalo + Toon + AdpativeSharpenGLSL:

DPIR(10)+DeHalo+Toon+ AdpativeSharpenGLSL:

DPIR(20)+DeHalo+Toon+ AdpativeSharpenGLSL:

DeHalo + Toon + Vsgan(2x_LD-Anime_Skr_v1.0):


-> there are tons of ways to clean it,... (and even more to change the colors, but I wouldn't try touching the colors without having a reference)

To decide how much denoising is okay, you need to look at a scene where you have strong optinion what are details and what not.

Cu Selur

Ps.: Sorry, I propably didn't always use the same frame, as I accidentially closed the viewer.
PPs.: @poisondeathray: "2x_LD-Anime_Skr_v1.0" is nice in combination with dehalo.

[lollo]

Yes, we all agreed on that. For this specific image we may say "noise" is removed, or "details" have been erased. Difficult to discern. Thanks for your samples!

[jagabo]

(Blu-ray source) due to the extreme black compression some scenes can be deemed unusable.

What about compositing the dark areas from other sources over the dark areas of the Blu-ray version? That's not too hard.

[poisondeathray]

(Blu-ray source) due to the extreme black compression some scenes can be deemed unusable.

What about compositing the dark areas from other sources over the dark areas of the Blu-ray version? That's not too hard.

Apparently they don't quite match up due to gate weave, slightly different characteristics

Good background info on the different versions here:
https://forum.doom9.org/showthread.php?p=1819724#post1819724

sephirotic:

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.

I guess they weren't super hardcore dedicated, only willing to spend days/weeks/months.

[rrats]

Am I missing something in the Twitter post? I see VLC screenshots, but no video sample.

They are enough to show off black compression aren't they?

[Selur]

At least not to me, since I don't use VLC and have no clue how accurate it is and I have no clue how the images were taken.
-> without a real video sample it's not really saying much for me

[rrats]

here

[Selur]

Ignoring the ghosting, here's what ChromaShiftSP + DeHalo_alpha + BasicVSR++

# Imports
import vapoursynth as vs
import os
import ctypes
# Loading Support Files
Dllref = ctypes.windll.LoadLibrary("i:/Hybrid/64bit/vsfilters/Support/libfftw3f-3.dll")
import sys
# getting Vapoursynth core
core = vs.core
# Import scripts folder
scriptPath = 'i:/Hybrid/64bit/vsscripts'
sys.path.insert(0, os.path.abspath(scriptPath))
# Loading Plugins
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/GrainFilter/RemoveGrain/RemoveGrainVS.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/GrainFilter/AddGrain/AddGrain.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/DenoiseFilter/NEO_FFT3DFilter/neo-fft3d.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/DenoiseFilter/DFTTest/DFTTest.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/Support/EEDI3m.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/ResizeFilter/nnedi3/NNEDI3CL.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/Support/libmvtools.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/Support/scenechange.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/Support/fmtconv.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/MiscFilter/MiscFilters/MiscFilters.dll")
core.std.LoadPlugin(path="i:/Hybrid/64bit/vsfilters/SourceFilter/d2vSource/d2vsource.dll")
# Import scripts
import chromashift
import havsfunc
# source: 'C:\Users\Selur\Desktop\demuxDVD.MPG'
# current color space: YUV420P8, bit depth: 8, resolution: 720x480, fps: 29.97, color matrix: 470bg, yuv luminance scale: limited, scanorder: top field first
# Loading C:\Users\Selur\Desktop\demuxDVD.MPG using D2VSource
clip = core.d2v.Source(input="E:/Temp/mpg_1e4c337e8e486dbb59f77d5c21ead8c9_853323747.d2v")
# Setting color matrix to 470bg.
clip = core.std.SetFrameProps(clip, _Matrix=5)
clip = clip if not core.text.FrameProps(clip,'_Transfer') else core.std.SetFrameProps(clip, _Transfer=5)
clip = clip if not core.text.FrameProps(clip,'_Primaries') else core.std.SetFrameProps(clip, _Primaries=5)
# Setting color range to TV (limited) range.
clip = core.std.SetFrameProp(clip=clip, prop="_ColorRange", intval=1)
# making sure frame rate is set to 29.970
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# setting field order to what QTGMC should assume (top field first)
clip = core.std.SetFrameProp(clip=clip, prop="_FieldBased", intval=2)
# Deinterlacing using QTGMC
clip = havsfunc.QTGMC(Input=clip, Preset="Fast", TFF=True, opencl=True) # new fps: 59.94
# make sure content is preceived as frame based
clip = core.std.SetFieldBased(clip, 0)
# Chroma adjustment using ChromaShiftSP
clip = chromashift.ChromaShiftSP(clip=clip, X=-2.00)
# cropping the video to 698x472
clip = core.std.CropRel(clip=clip, left=12, right=10, top=4, bottom=4)
clip = havsfunc.DeHalo_alpha(clip)
clip = core.std.AddBorders(clip=clip, left=0, right=6, top=0, bottom=0) # add borders to archive mod 8 (vsBasicVSRPPFilter) - 704x472
# adjusting color space from YUV420P8 to RGBS for vsBasicVSRPPFilter
clip = core.resize.Bicubic(clip=clip, format=vs.RGBS, matrix_in_s="470bg", range_s="limited")
# Quality enhancement using BasicVSR++
from vsbasicvsrpp import BasicVSRPP
clip = BasicVSRPP(clip=clip, model=3)
clip = core.std.CropRel(clip=clip, left=0, right=6, top=0, bottom=0) # removing borders (vsBasicVSRPPFilter) -  698x472
# adjusting output color from: RGBS to YUV420P10 for x265Model
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P10, matrix_s="470bg", range_s="limited", dither_type="error_diffusion")
# set output frame rate to 59.940fps
clip = core.std.AssumeFPS(clip=clip, fpsnum=60000, fpsden=1001)
# Output
clip.set_output()

do to the demuxDVD.MPG file
I see the issue with the black levels in 00008_cutted.m2ts, damn they removed the ghosting but made a mistake when mapping the colors. :/