Is there any software out there that automatically converts Black and White

[Bernix]

In future I believe there will be soft that using deep learning, object recognition, motion tracking, face detection (skin detection) will speed up the colorizing process. But it will needed huge database of object and colors. And there are objects you can't assign (pullover, walls, roofs, etc....) Not only tank is green, but its exactly green shade. So it will be much much faster, but human supervision will still be needed, less but needed.

[raffriff42]

It's been talked about for a long long time...

As near as I can tell each of your links discusses recovering the color from sources originally shot in color. This is different from the OP's question about colorization of black and white sources...

Ah, sorry, you're right.

Is there software that can do this kind of processing now?

https://forum.doom9.org/showthread.php?t=173364

I have played with the AviSynth script posted in this thread and it kinda works, or at least produces sometimes-interesting results. R.STEED should take a look at it. The worst that can happen is he learns to use AviSynth.

(The author of this script initially claimed he could recover color information from the grayscale + "the frequencies" but had to retract / delete that claim after he was politely destroyed)

[jagabo]

There is another class of color restoration where chroma dots on greyscale film are converted to color. This only works for a particular class of film: that which was made by filming a television. In the early days of color video tape the tape was very expensive. So after a TV episode had aired it was common practice to re-use the tape for another show. Typically, the BBC would convert the video to b/w film by pointing a film camera at a TV. These less expensive films were then sent to secondary markets. It was discovered that enough of the chroma subcarrier was visible on the film that colors could be restored from them.

https://www.youtube.com/watch?v=CjK-b4x9ZmQ

A 1:58 into the the clip you can see an enlargement where the chroma dots are visible.

[Bernix]

There are two ways. 1. Dad's army, which was filmed in color but only B&W recording survived. Here is the result. It contains color information even in B&W. But filmed as B&W cant be recover this way. And this is only Real color restoring method. https://www.youtube.com/watch?v=CjK-b4x9ZmQ
2. There are machine deep learning method. It can make color from B&W footage but quality now isnt best. It is based on many training images. It is good for sky, tree and watter and ground and faces.

Bernix

Jagabo, I am on your ignore list

Bernix

[jagabo]

Jagabo, I am on your ignore list

No, I just missed the post.

[MWilson]

(The author of this script initially claimed he could recover color information from the grayscale + "the frequencies" but had to retract / delete that claim after he was politely destroyed)

Haha More like annoyed.
After a lot of effort I've got a script that can colorize video sufficiently 60 ~ 70 % of the time (I know you would like to argue this, but if you get a halfway decent colorization with no manual intervention, and pay absolutely nothing for it, how is that a loss?).

[Bernix]

MWilson,
are you coloring B&W old movies with this script?

Bernix

[MWilson]

MWilson,
are you coloring B&W old movies with this script?

Bernix

Yes I am.

[Bernix]

o.k. so how do you know the colors are right. Try colorize New York taxi car (cab?) by this method and let me know.
And maybe I surprised you, this is excellent result, but mainly because skin detection and grass detection. This is results of my childhood
But you can notice other colors are false. There is skin tone on leg, but it is not skin. But still good. The mushroom and the color of shirt is also false, almost everything except grass and trees and skin. And of course white is white and black is black.

Bernix

[MWilson]

o.k. so how do you know the colors are right. Try colorize New York taxi car (cab?) by this method and let me know.
And maybe I surprised you, this is excellent result, but mainly because skin detection and grass detection. This is results of my childhood
But you can notice other colors are false. There is skin tone on leg, but it is not skin. But still good. The mushroom and the color of shirt is also false, almost everything except grass and trees and skin. And of course white is white and black is black.

Bernix

60~70% of the time the results are (what I would assume to be) sufficient.

Let me put this out there once again

If you get a halfway decent colorization with no manual intervention, and pay absolutely nothing for it, how is that a loss?

Food for thought

[Bernix]

No food for thought,
if something like this is for you enough.
Check original, black and white and colorized.
I bet you have similar results....
EDIT this is much better example


Bernix

[Bernix]

New, to don't missed it http://hi.cs.waseda.ac.jp:8082/
I just uploaded black to white transition picture, and it throw away blue to white, thinks it is sky probably.
All this computer colorization are nice toys, but color (except sky, water, skin, and trees and grass) is just guessing. If this is ok, everything is o.k.

Bernix

[MWilson]

No food for thought,
if something like this is for you enough.
Check original, black and white and colorized.
I bet you have similar results....
EDIT this is much better example


Bernix

Over smoothed clip ran through the latest script.
http://colorize.dev.kaisou.misosi.ru/result/983195ce-58c4-4539-879a-2bb3a6514a90

I would have to say the script did better.

[Bernix]

So post results from the script, what you posted isnt good, almost 50% of the picture is sky, mentioned earlier. Try the parking cars I posted in your script and show result.

Bernix

[MWilson]

[Attachment 43578 - Click to enlarge]

Sorry for the picture of my screen, I'm not home, but I had already posted this pic.

When I get home, I'll be happy to show you something on your pic (Although what I said a couple posts up is honestly not changing).

[DB83]

I must be missing something here.

I held my breath and went to that Russian link

Firstly the grey-scale is not original B+W. Even then there is little color (or even colour) in the result. And most of that is wrong

[MWilson]

I must be missing something here.

I held my breath and went to that Russian link

Firstly the grey-scale is not original B+W. Even then there is little color (or even colour) in the result. And most of that is wrong

Yep. A lot of people hate my scripts for colorization and are perfectly satisfied with any neural network.



Pot. Kettle. Black.

[DB83]

^^ What is neural ?

And where have you seen me attempt any colorization ?

Unless you also do not know the correct usage of the idiom you have quoted.

Hmmn. Wasn't Jon Pertwee's Doctor Who in colour to start with ? Or is that just another crass attempt with your pet ? If it is you have created even more problems than you solve.

[Bernix]

Hi,
you can choose japan url from original developers. It accepted both Black and white, but color also, it desaturate it first, so you can compare original and result. I use this to show, especially on the picture, where are parking cars, that it simply guessing colors (for most thing). Therefore original B&W and result.
Edit: Sorry for not knowing english idioms
Edit:http://hi.cs.waseda.ac.jp:8082/
Japan link to original developers of this. Something looks good, but most dont. Of course in gallery, there is best results
Bernix

[MWilson]

^^ What is neural ?

And where have you seen me attempt any colorization ?

Unless you also do not know the correct usage of the idiom you have quoted.

Hmmn. Wasn't Jon Pertwee's Doctor Who in colour to start with ? Or is that just another crass attempt with your pet ? If it is you have created even more problems than you solve.

Wow, you didn't get anything I just said.
First, who on earth said anything about you colorizing anything. Second, yes it was color to start with, I desaturated the video and ran through the script. Get this straight, if you want to point out inconsistencies in my scripts, and not put the same amount of scrutiny to any other method - That's the pot calling the kettle black (Neither is perfect).

I must be missing something here.

Yep.

[Bernix]

interesting reading http://hi.cs.waseda.ac.jp/~iizuka/projects/colorization/data/IizukaSIGGRAPH2016_slide.pdf
comparison of method http://hi.cs.waseda.ac.jp/~iizuka/projects/colorization/en/
with video of lost world 1925 and living dead also old.
But it is still guessing.

Bernix

[DB83]

Nope. Just learn some proper English or look it up.

You decided to troll this topic with your script. So any criticism is valid.

But you have done the topic one great service by disproving the OPs argument of each shade of grey-scale having an equivalent color. Although that opinion had already been rubbished earlier.

Now if you want real criticism just post a colourized video from a genuine B+W source.

[MWilson]

You decided to troll this topic with your script. So any criticism is valid.

Yep, I decided to

(The author of this script initially claimed he could recover color information from the grayscale + "the frequencies" but had to retract / delete that claim after he was politely destroyed)

troll.

------

Clarification : I'm sick of people insulting me, my intelligence, or both.

[DB83]

Soldiers learnt quite quickly that if they stuck their heads over a paparpet they would get shot in the head.

Or. If you can not take valid criticism you should avoid the post.

Find the real useage of Pot-Kettle yet ?. You might wish to accuse me of bias but that is a whole different ball-game.

[MWilson]

Soldiers learnt quite quickly that if they stuck their heads over a paparpet they would get shot in the head.

Or. If you can not take valid criticism you should avoid the post.

Find the real useage of Pot-Kettle yet ?. You might wish to accuse me of bias but that is a whole different ball-game.

Okay here goes. Without using any clues from my scripts , come up with a way to colorize video in avisynth and have it look even half decent. Post your results often , see what kind of crap you’re getting each day. When you finally have something to be halfway proud of, watch people rip into it and you for no good reason. <— That is valid criticism.

[DB83]

I do not have to rise to that challenge. The whole point of a forum is to react to what you read. And as the saying goes "If you can not take the heat then get out of the kitchen"

And I am reacting to what I see in Post #43. The only part of the result that is even 'Half Decent' (and these words are subjective) is the sky. And most know that sky is usually blue. But if sky occupies half of your picture you might argue that the end result is half decent. Except that I doubt you meant that.

As for the rest of that picture. Do tell me where the color is. Then I will book an appointment with my friendly optician.

[Bernix]

Sorry, but I am bit confused. The page, you posted that is your script, the blurry picture i don't understand. That web service is provided by user here Mecab. Why did you changed your nick? That for first. And second. From the page

I am not related to the original project directly nor indirectly. For questions, suggestion, or any other comments regarding this web app, please let me know. (Appreciations should go to the original authors..., in the backend, the server just runs their code simply ) (Oct. 30, 2016) Original authors have released the colorization web service officially!

So you are Mecab or one of these Satoshi Iizuka, Edgar Simo-Serra or Hiroshi Ishikawa ?
At first look I missed that you claimed the blurry picture is made by your script! I thought it comes later. So I am really confused now. But I am missing something for sure, as always. There will be misunderstood on my side. BTW my pictures used same service as your blurry picture.

Bernix

[Cornucopia]

OK, here's a "put up or shut up" moment:
Attached is a picture that has six stripes of (completely) different colors upon a white background. It has been desaturated to greyscale. They all basically have the same grey level. Restore the colors! I've saved both, so when the time comes, I'll show you the original.

Scott


[Attachment 43579 - Click to enlarge]

(edit: note I've already given you more "context" and clues that many images would give you. only 6 (+1) colors should be simple)

[Bernix]

OK, here's a "put up or shut up"

my previously idea about test was much simpler. Who is Huey, Dewey, and which is Louie. It is simple because of different shades of gray.
But still curious. Even it is possible find out by episode etc. I could do it bit harder not desaturate, but using B&W film emulation, which result in very different shades of gray. But just for curiosity

[Attachment 43580 - Click to enlarge]
I wish no cheating in searching for episode and desaturate them.
So it isnt real put up shut up test

[DB83]

I have just watched an extra on a 2-disk version of Holiday Inn - Disk 1 is the original B+W. Disk 2 is the colorized version.

Whilst I would still prefer to watch Disk 1, the results shown in the extra are really impressive. They mention 'programs' but do not go in to any real detail other than the whole project took 6 weeks.

Mentioned earlier. The docu does reveal that all the hard work was done in India. Purely, I guess, for reasons of cost.