What if they had colors back then?
August 13, 2016 9:40 AM Subscribe
In fact, those old photographs are in color. It's just the world was black and white then
posted by obscure simpsons reference at 11:05 AM on August 13, 2016 [12 favorites]
posted by obscure simpsons reference at 11:05 AM on August 13, 2016 [12 favorites]
You too can be just like Ted Turner!
posted by jim in austin at 11:29 AM on August 13, 2016 [3 favorites]
posted by jim in austin at 11:29 AM on August 13, 2016 [3 favorites]
why?
posted by philip-random at 11:43 AM on August 13, 2016 [3 favorites]
posted by philip-random at 11:43 AM on August 13, 2016 [3 favorites]
I was curious about how good a job this algorithm did of recreating actual colors rather than just plausible colors, so Indid a little test by taking a well-known image, one where color plays a major role, desaturating it, and then running it through the bot. I ended up going with "Afghan Girl," though I'd be curious to see how the experiment goes with other photos.
Original
Desaturated
Recolorized
Interesting results. Closer, actually, than I expected. Not that close, but not as bad as I had assumed. Anyone else want to play?
posted by Anticipation Of A New Lover's Arrival, The at 11:59 AM on August 13, 2016 [5 favorites]
Original
Desaturated
Recolorized
Interesting results. Closer, actually, than I expected. Not that close, but not as bad as I had assumed. Anyone else want to play?
posted by Anticipation Of A New Lover's Arrival, The at 11:59 AM on August 13, 2016 [5 favorites]
Hate colorising. Leave them in glorious monochrome. What is it supposed to be adding?
posted by GallonOfAlan at 1:34 PM on August 13, 2016 [3 favorites]
posted by GallonOfAlan at 1:34 PM on August 13, 2016 [3 favorites]
Thinking about it a little more, the picture loses pretty much everything that makes it iconic after its run through the algorithm. There's no contrast of red and green, and the subject's eyes aren't nearly as hypnotic.
I'm sort of in the "interesting proof of concept, but why?" camp. Obviously people have been colorizing monochrome photos for almost as long as photography has existed, but there's typically been some authorial intent behind it, a desire to recreate reality or to achieve a particular effect. Without that, I'm not sure what colorization actually adds to the picture.
I'm also concerned (in a matter-of-principle sort of way, not a this-is-actually-a-problem way) that replacing a lack of color in a historical photo with an algorithm's guess as to what some plausible colors might have been is sort of a distortion of history. I'm having a hard tome articulating exactly why I have an issue with this, but something about it rubs me the wrong way. Something something biased algorithms something.
posted by Anticipation Of A New Lover's Arrival, The at 1:38 PM on August 13, 2016 [4 favorites]
I'm sort of in the "interesting proof of concept, but why?" camp. Obviously people have been colorizing monochrome photos for almost as long as photography has existed, but there's typically been some authorial intent behind it, a desire to recreate reality or to achieve a particular effect. Without that, I'm not sure what colorization actually adds to the picture.
I'm also concerned (in a matter-of-principle sort of way, not a this-is-actually-a-problem way) that replacing a lack of color in a historical photo with an algorithm's guess as to what some plausible colors might have been is sort of a distortion of history. I'm having a hard tome articulating exactly why I have an issue with this, but something about it rubs me the wrong way. Something something biased algorithms something.
posted by Anticipation Of A New Lover's Arrival, The at 1:38 PM on August 13, 2016 [4 favorites]
"interesting proof of concept, but why?"
Is this question not blatantly self-answering? I don't think the image object is the focus here.
posted by invitapriore at 3:29 PM on August 13, 2016 [3 favorites]
Is this question not blatantly self-answering? I don't think the image object is the focus here.
posted by invitapriore at 3:29 PM on August 13, 2016 [3 favorites]
I don't know why anyone would think that this applies to historic photos. I just want it for things that I've accidentally saved in black and white after ditching the original, or images from the internet that i need for reference that i can only find in b&w. i'm sure this can and will work someday, but, yeah this doesn't really work that well.
posted by Conrad-Casserole at 3:57 PM on August 13, 2016
posted by Conrad-Casserole at 3:57 PM on August 13, 2016
This also almost certainly assists in future attempts to interpret b&w pictures of things, for machine learning
posted by DoctorFedora at 4:15 PM on August 13, 2016
posted by DoctorFedora at 4:15 PM on August 13, 2016
I think you'd have to be pretty jaded not to be impressed by this.
posted by uosuaq at 5:54 PM on August 13, 2016 [3 favorites]
posted by uosuaq at 5:54 PM on August 13, 2016 [3 favorites]
Those are a couple of pretty good reasons that I hadn't thought of, so thanks.
posted by Anticipation Of A New Lover's Arrival, The at 6:39 PM on August 13, 2016
posted by Anticipation Of A New Lover's Arrival, The at 6:39 PM on August 13, 2016
Hate colorising. Leave them in glorious monochrome. What is it supposed to be adding?
I've been following this woman on twitter and facebook. I think she colorizes by hand, and the results are uncanny.
posted by stargell at 7:59 PM on August 13, 2016 [14 favorites]
I've been following this woman on twitter and facebook. I think she colorizes by hand, and the results are uncanny.
posted by stargell at 7:59 PM on August 13, 2016 [14 favorites]
Thanks stargell. Marina Amaral's work is phenomenal. Goes to show that a classic BW is a classic colour.
I'm less convinced that the other way around is as effective; its hard to unsee the colours once seem...but, then again...BW is like memory....
posted by GhostRider at 8:22 PM on August 13, 2016
I'm less convinced that the other way around is as effective; its hard to unsee the colours once seem...but, then again...BW is like memory....
posted by GhostRider at 8:22 PM on August 13, 2016
I'm less convinced that the other way around is as effective; its hard to unsee the colours once seem.
Really? I convert pictures to B&W all the time.
I don't know if it's nature or nurture but there definitely seems to be a preference for most people.
posted by bongo_x at 10:14 PM on August 13, 2016
Really? I convert pictures to B&W all the time.
I don't know if it's nature or nurture but there definitely seems to be a preference for most people.
posted by bongo_x at 10:14 PM on August 13, 2016
This is an impressive demonstration of machine learning. As to its practical utility as a photo processing tool rather than an AI testing tool, there are evidently a large number of people who want to colourise monochrome photos, for whatever reason, and at least some of those would be happy with at least some of these results. So there's that.
I think the Afghan Girl photo is a poor example because it sets the wrong expectations. The colour information is literally absent from a monochrome photo and there is no way to recover the actual data except by guessing based on other knowledge. Thus the algorithm was able to work out that the face was a skin-tone and so reconstructed the colour almost exactly. That's possible because all skin colours are essentially on a single dimension from light to dark so all the information you need is in a monochrome photo. This reconstruction is, however, founded on the assumption that the subject is not, e.g., wearing green face-paint. Note that the same thing does not apply to eye colours and thus the reconstruction is not as good. Unfortunately one of the most important aspects of this photo is, as noted, the colour component of the foreground/background contrast and there is no way to reconstruct that from the monochrome image directly. That said, the photo is so well known that the AI could conceivably have the original in its knowledge corpus and thus reconstruct it exactly. Would that be cheating?
The corruption-of-historical-record aspect of all this is interesting but a drop in the ocean of what historians already have to deal with, I suspect. There is simply no such thing as an accurate historical record of anything, life being as subjective as it is.
posted by merlynkline at 11:07 PM on August 13, 2016 [2 favorites]
I think the Afghan Girl photo is a poor example because it sets the wrong expectations. The colour information is literally absent from a monochrome photo and there is no way to recover the actual data except by guessing based on other knowledge. Thus the algorithm was able to work out that the face was a skin-tone and so reconstructed the colour almost exactly. That's possible because all skin colours are essentially on a single dimension from light to dark so all the information you need is in a monochrome photo. This reconstruction is, however, founded on the assumption that the subject is not, e.g., wearing green face-paint. Note that the same thing does not apply to eye colours and thus the reconstruction is not as good. Unfortunately one of the most important aspects of this photo is, as noted, the colour component of the foreground/background contrast and there is no way to reconstruct that from the monochrome image directly. That said, the photo is so well known that the AI could conceivably have the original in its knowledge corpus and thus reconstruct it exactly. Would that be cheating?
The corruption-of-historical-record aspect of all this is interesting but a drop in the ocean of what historians already have to deal with, I suspect. There is simply no such thing as an accurate historical record of anything, life being as subjective as it is.
posted by merlynkline at 11:07 PM on August 13, 2016 [2 favorites]
« Older Only shooting stars break the mold | Screwed up eyes and screwed down hairdo Newer »
This thread has been archived and is closed to new comments
Still, this is an amazing leap forward considering a few years ago there was no ability for machines to parse images at all.
posted by blahblahblah at 10:52 AM on August 13, 2016