Color Correcting the Dragon Box - 3 Part Spectacular

Forte224 · Post by **Forte224** » Sun Apr 14, 2019 9:55 pm

JohnnyCashKami wrote: ↑Sun Apr 14, 2019 9:51 pm Just for the fun of it, I tried to fix the colors on TV SP2 which looked a bit too green and here's the result.

Original
Spoiler:
CC'd

Spoiler:

Did this via my Samsung tablet with LightX. Looks better to me without that green tint or whatever it is.

Green tint might be TOEI's "DRM" which everyone despise but accept regardless.

Looks a bit dark but otherwise nice. Someone really needs to figure out that green tint issue at Toei. Unless it IS intentional. Which is just....ugh.

lansing · Post by **lansing** » Mon Apr 15, 2019 3:51 am

Forte224 wrote: ↑Sun Apr 14, 2019 9:55 pm
JohnnyCashKami wrote: ↑Sun Apr 14, 2019 9:51 pm Just for the fun of it, I tried to fix the colors on TV SP2 which looked a bit too green and here's the result.

Original
Spoiler:
CC'd

Spoiler:

Did this via my Samsung tablet with LightX. Looks better to me without that green tint or whatever it is.

Green tint might be TOEI's "DRM" which everyone despise but accept regardless.
Looks a bit dark but otherwise nice. Someone really needs to figure out that green tint issue at Toei. Unless it IS intentional. Which is just....ugh.

I had figured it out last year, the answer is color matching. Just start reading from 10 page back in this thread and you'll see. It was kind of a testing phase last year as many things were still unclear. But my knowledge is pretty solid now. Hopefully I can get this whole thing working and the vision is, if this works on dragon ball, it will work on all other old anime.

This is color matching the last image to a cel image of Gohan I have. Notice the bean is gray out because I don't have a reference color for it.

Spoiler:

Forte224 · Post by **Forte224** » Mon Apr 15, 2019 8:43 am

lansing wrote: ↑Mon Apr 15, 2019 3:51 am
Forte224 wrote: ↑Sun Apr 14, 2019 9:55 pm
JohnnyCashKami wrote: ↑Sun Apr 14, 2019 9:51 pm Just for the fun of it, I tried to fix the colors on TV SP2 which looked a bit too green and here's the result.

Original
Spoiler:
CC'd

Spoiler:

Did this via my Samsung tablet with LightX. Looks better to me without that green tint or whatever it is.

Green tint might be TOEI's "DRM" which everyone despise but accept regardless.
Looks a bit dark but otherwise nice. Someone really needs to figure out that green tint issue at Toei. Unless it IS intentional. Which is just....ugh.
I had figured it out last year, the answer is color matching. Just start reading from 10 page back in this thread and you'll see. It was kind of a testing phase last year as many things were still unclear. But my knowledge is pretty solid now. Hopefully I can get this whole thing working and the vision is, if this works on dragon ball, it will work on all other old anime.

This is color matching the last image to a cel image of Gohan I have. Notice the bean is gray out because I don't have a reference color for it.

Spoiler:

That's gorgeous.

HakkaiBills93 · Post by **HakkaiBills93** » Wed Apr 17, 2019 6:21 am

lansing wrote: ↑Mon Apr 15, 2019 3:51 am
Forte224 wrote: ↑Sun Apr 14, 2019 9:55 pm
JohnnyCashKami wrote: ↑Sun Apr 14, 2019 9:51 pm Just for the fun of it, I tried to fix the colors on TV SP2 which looked a bit too green and here's the result.

Original
Spoiler:
CC'd

Spoiler:

Did this via my Samsung tablet with LightX. Looks better to me without that green tint or whatever it is.

Green tint might be TOEI's "DRM" which everyone despise but accept regardless.
Looks a bit dark but otherwise nice. Someone really needs to figure out that green tint issue at Toei. Unless it IS intentional. Which is just....ugh.
I had figured it out last year, the answer is color matching. Just start reading from 10 page back in this thread and you'll see. It was kind of a testing phase last year as many things were still unclear. But my knowledge is pretty solid now. Hopefully I can get this whole thing working and the vision is, if this works on dragon ball, it will work on all other old anime.

This is color matching the last image to a cel image of Gohan I have. Notice the bean is gray out because I don't have a reference color for it.

Spoiler:

Dr dre tools is really impressive, i hope he will make a new version of it for example for the small areas colors
by the way what i gave you aren't production cels of course but skin tone etc aren't nice on them? can help for some colors
the only issue that prevent me using dr dre tools is that the lut that is build with color match don't give me the right issue when i import it to after effect , when you'll be done i hope that a kind of tutorial or preset could be done for noob like me^^

gokaiblue · Post by **gokaiblue** » Fri Apr 19, 2019 2:24 pm

Did anyone ever post an updated guide on this?

crabshank1 · Post by **crabshank1** » Wed May 01, 2019 12:08 am

As I posted earlier, a correct white balance is one where for all pixels that are supposed to be greyscale (including white and black) R=G=B (saturation=0 in HSV).

So I copied and pasted a few areas from screenshots of one episode of The Final Chapters that I thought should be grey into one image:

then I ran some code I wrote which matches the RGB channels to the Value channel (0-255) of the image and then remapped each channel accordingly and got this:

Original:

Remapped:

http://www.framecompare.com/screenshotc ... n/G76PNNNX

crabshank1 · Post by **crabshank1** » Tue May 28, 2019 5:34 pm

crabshank1 wrote: ↑Wed May 01, 2019 12:08 am As I posted earlier, a correct white balance is one where for all pixels that are supposed to be greyscale (including white and black) R=G=B (saturation=0 in HSV).

So I copied and pasted a few areas from screenshots of one episode of The Final Chapters that I thought should be grey into one image:

then I ran some code I wrote which matches the RGB channels to the Value channel (0-255) of the image and then remapped each channel accordingly and got this:

Original:

Remapped:

http://www.framecompare.com/screenshotc ... n/G76PNNNX

The tools I'm using, if anyone's interested: https://github.com/crabshank/HLSL-Resha ... -remappers

ect5150 · Post by **ect5150** » Thu May 30, 2019 8:02 am

HakkaiBills93 wrote: ↑Fri Apr 12, 2019 5:49 am

In the bottom right corner, COLOR of XEROX LINE. RGB [15,13,15]. That's as accurate of a reference point for anything as you'd get for something like this. Question is... what's the XEROX LINE? Either way, I'm guessing that blackbox in the bottom right is supposed to be [15,13,15] and we can use that to make certain the other boxes are accurate for the eyes and hair.

Did you spot others like this?

crabshank1 · Post by **crabshank1** » Thu May 30, 2019 5:39 pm

I know sharpening's a contentious area, but I trained a neural network to recognise lines and ported it to HLSL (and added a bit of maths to make all the lines black) so it could work on realtime video:

Then you darken the areas that are black in the black and white image.

ect5150 · Post by **ect5150** » Fri May 31, 2019 8:51 am

crabshank1 wrote: ↑Tue May 28, 2019 5:34 pm The tools I'm using, if anyone's interested: https://github.com/crabshank/HLSL-Resha ... -remappers

Did you code these tools? I tried playing around with around with histogram and histogram neuralnet pages.

I just ask because I'm having a few issues with it and might want to help with a little coding. I'd like to make a few adjustments and have it be used with some of the software mentioned a few pages back (at the start of the guide). Especially if the hue mapping is highly accurate with the non-neuralnet.

Also, that neuralnet histogram is painfully slow. I just shut it down before it stopped.

Using the following images to play with it, mapping the FUNI Dragon Ball sets to DB DBox. But I want to be able to take the results and put it into VirtualDub or AVISynth (or some other free software) to use.

Spoiler:

It currently generated the following info for the hue mapping. I'm working on putting into VirtualDub. Fairly certain I'll have it by the end of the day (work calls first). We'll see how well it looks and then I'll play around with the other stuff (Saturation and Value too).

Spoiler:

Code: Select all

crabshank1 · Post by **crabshank1** » Fri May 31, 2019 4:54 pm

ect5150 wrote: ↑Fri May 31, 2019 8:51 am
crabshank1 wrote: ↑Tue May 28, 2019 5:34 pm The tools I'm using, if anyone's interested: https://github.com/crabshank/HLSL-Resha ... -remappers
Did you code these tools? I tried playing around with around with histogram and histogram neuralnet pages.

I just ask because I'm having a few issues with it and might want to help with a little coding. I'd like to make a few adjustments and have it be used with some of the software mentioned a few pages back (at the start of the guide). Especially if the hue mapping is highly accurate with the non-neuralnet.

Also, that neuralnet histogram is painfully slow. I just shut it down before it stopped.

Using the following images to play with it, mapping the FUNI Dragon Ball sets to DB DBox. But I want to be able to take the results and put it into VirtualDub or AVISynth (or some other free software) to use.

Spoiler:

Did you code these tools?

I did code them myself.

I just ask because I'm having a few issues with it and might want to help with a little coding.

If you want/can you can port rgb_hsv match.hlsl to VirtualDub or AVISynth.

Also, that neuralnet histogram is painfully slow.

You have to go into the code and adjust the hyperparameters (learning rate, no. of layers etc.), and also smaller images work better e.g. these were my training images for the line detection:

The problems with the hues in these images were that there weren't many different ones and I had to map 357° -> 14°, so I had to rotate the hues first (my web page can't deal with hue rotation, but my shader can). In the end I just rotated the hues and remapped the values, as remapping the saturations caused artefacts. Also the images weren't exactly the same frame, but here're my results:

Dragon Box

Remapped Dragon Box

Funi sets

http://www.framecompare.com/screenshot ... n/MMEJNNNU

Robo4900 · Post by **Robo4900** » Fri May 31, 2019 6:02 pm

crabshank1 wrote: ↑Fri May 31, 2019 4:54 pm
Remapped Dragon Box

That's way, way, way more orange than it should be.

crabshank1 · Post by **crabshank1** » Fri May 31, 2019 6:04 pm

Robo4900 wrote: ↑Fri May 31, 2019 6:02 pm
crabshank1 wrote: ↑Fri May 31, 2019 4:54 pm
Remapped Dragon Box
That's way, way, way more orange than it should be.

I just matched the hue from the Funi sets, it's more a proof of concept, I could do any hue.

crabshank1 · Post by **crabshank1** » Fri May 31, 2019 6:07 pm

Spoiler:

Code: Select all

I suggest using the recommended remappings as they are designed to capture the most of the full remappings in the fewest number of points whilst making the spaces between them as even as possible.

TheBalishChannel · Post by **TheBalishChannel** » Fri May 31, 2019 8:00 pm

I tried slamming in a shot from Episode 86: The Last Wish into a program called "Topaz Gigapixel." The point of the program is to artificially enhance the detail in any photograph (or in this case, a screenshot) with "machine learning." In my opinion, it doesn't look too bad for being done through an automated process.

Click here to view a comparison.

crabshank1 · Post by **crabshank1** » Fri May 31, 2019 9:35 pm

TheBalishChannel wrote: ↑Fri May 31, 2019 8:00 pm I tried slamming in a shot from Episode 86: The Last Wish into a program called "Topaz Gigapixel." The point of the program is to artificially enhance the detail in any photograph (or in this case, a screenshot) with "machine learning." In my opinion, it doesn't look too bad for being done through an automated process.

Click here to view a comparison.

Was the original really that macroblocked?

ect5150 · Post by **ect5150** » Sat Jun 01, 2019 10:54 am

crabshank1 wrote: ↑Fri May 31, 2019 4:54 pm The problems with the hues in these images were that there weren't many different ones and I had to map 357° -> 14°, so I had to rotate the hues first.

I noticed the same thing after the fact. Also, it seemed to work "well enough" for that single frame... but in the very next scene, Piccolo's face was front and center and the color was WAY off (looks like it's just a sampling piece there).

Robo4900 wrote: ↑Fri May 31, 2019 6:02 pm That's way, way, way more orange than it should be.

As was previously stated, it was just remapping the hue. It's probably possible to do the same with the saturation and value information as well.

Ideally, maybe we could setup a script that would accept two videos (a source and a video to alter... FUNI's box and the Dragonbox), and it would run this process frame by frame. There are some minor frame differences between the DragonBox and FUNI's boxes, but if we could pull this off, it would go a long way to automating large portions of it.

ect5150 · Post by **ect5150** » Sat Jun 01, 2019 10:58 am

TheBalishChannel wrote: ↑Fri May 31, 2019 8:00 pm I tried slamming in a shot from Episode 86: The Last Wish into a program called "Topaz Gigapixel." [snip]

Spoiler:

Click here to view a comparison.

That's the best upscale of anything I've ever seen.

crabshank1 · Post by **crabshank1** » Sat Jun 01, 2019 10:01 pm

As was previously stated, it was just remapping the hue. It's probably possible to do the same with the saturation and value information as well.

I did the values too for that one, it's quite subtle.

Ideally, maybe we could setup a script that would accept two videos (a source and a video to alter... FUNI's box and the Dragonbox), and it would run this process frame by frame. There are some minor frame differences between the DragonBox and FUNI's boxes, but if we could pull this off, it would go a long way to automating large portions of it.

This is doable and worth pursuing, just need to figure out to which programs to port my code.

crabshank1 · Post by **crabshank1** » Sun Jun 02, 2019 4:55 pm

Ideally, maybe we could setup a script that would accept two videos (a source and a video to alter... FUNI's box and the Dragonbox), and it would run this process frame by frame. There are some minor frame differences between the DragonBox and FUNI's boxes, but if we could pull this off, it would go a long way to automating large portions of it.

Scene-by-scene should be enough. Porting my shader to .avs will be a massive step forward, but I've never worked with .avs before

.

Kanzenshuu

Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Re: Color Correcting the Dragon Box - 3 Part Spectacular

Who is online