If you're talking about "restoring" in the sense of fixing the holes: I doubt ML can do anything when a huge section of the picture is missing. Any detail you put in would be hallucinated and not representative of reality.
Sure, but Photoshop (for example) has a feature that does just that: Content-Aware Fill. https://helpx.adobe.com/photoshop/using/content-aware-fill.h...
A good algorithm will absolutely represent reality, otherwise you would notice that it's not working. It may not reflect the actual reality behind the black hole, but it would remove the whole, making it indistinguishable from what you would think should be behind the hole.
With claims of "the hole is part of the history" aside, it would make these photos much easier to look at, allowing one to focus on the content of the photo, rather than the hole.
There is no algorithm in existence or will ever be discovered in our lifetimes that could cover those black holes with enough accuracy, or creativity, to make it indistinguishable from the real thing.
And therein lies the problem. I'd rather see literally nothing, a black hole, than some AI-constructed representation of what an algorithm determines 'should' be there.
And regardless of whether or not you want to call it censorship or some obfuscated term like "editorial choices", the point remains that the black holes draw attention to these "editorial choices", which seems to very clearly be the point of this project. Removing them, or making them unobtrusive is counter to the aims of the project.
See: https://news.ycombinator.com/item?id=30281686
After verifying this with around 20 random photos, I now agree with that comment, and disagree with the article. The article doesn't seem to have merit.
If you define "represent reality" as "lie in a convincing way", then maybe. But, it has zero with reality.
But that is likely to mislead people into thinking that what filled the hole is what was really there at the time.
Even small errors can radically change an image, as this (super-resolution rather than content-aware-fill) example shows: https://www.theverge.com/21298762/face-depixelizer-ai-machin...
I don't think this is a good metric. For fun and entertainment, it works, but lots of contexts need more than that. Take an example of lossy audio, like MP3. People are generally fine with 128k mp3s, but I doubt audio engineers would want to store masters in that format and bitrate.
In case of colorization, the issue is that the color really is lost on the film. Multiple shades of different colors map to the same shade of grey on black and white media - my favorite example is that B&W TV make-up was bright green. So them no matter what you do, it'll be just a conjecture of what the original color might have been.
Don't buy the "censored" garbage the article is peddling, it's not true.
Some of the photos are obviously flawed or uninteresting, but others look good and even touching. Why ruin them?
Presumably if there was an entire strip of bad ones, it would be tossed, but most of the time it was a mix, and it was a way to mark them.
Not a great way, since someone else might like the photo, but I guess he saw zero value in the poor photos (basically the same as pressing delete on a digital camera - do you keep every photo you take, even bad ones?)
Don't forget - he wasn't working off of "historical" photos, to him, these were present day photos, and easily replaceable.
To us these are irreplaceable parts of history, but it wasn't like that to him.
> It's still sad that the censor
Editor, not censor - he was preparing interesting photos for publication.
Unfortunately yes. This is the problem with digital photos, at least on a phone: the effort to remove the bad ones is not worth it for someone like me who takes pictures occasionally and never looked at them anyway (this is why I do not take a lot either)
I wonder if people who are not into photography actually manage their photos (sort, tag, prune)?
The last year I took a camera again, with a mobile photo too. After a couple of hundreds of shots and a futile attempt to manage them I understood what while some photos could be of interest for me a couple decades later, if I don't have a strong desire to save some exact shots now - I just wipe them all.
Of course I found myself wanting to show a couple of photos half a year after I wiped them, but the value was in showing them to a person I unexpectedly met again, not in the photos themselves.
I still don't touch my old library. "Some other day".
Though if I manage to organize and merge my two NextCloud instances I would probably upload some of the photos to it for the safekeeping.
> basically the same as pressing delete on a digital camera - do you keep every photo you take, even bad ones?
I used to keep everything, now I make an effort to delete the crap or even redundant photos. I agonize over this, but I try to do it. Then again, most photos I take I never look at again in my life, which I'm guessing happens to many of us in the age of digital cameras.
> To us these are irreplaceable parts of history, but it wasn't like that to him.
Yes, this is the crux of the problem.
Like you, I don't think this was censorship. I used the word "censor" because that's how the article frames it, but I agree this was an act of editing/choosing, not censorship.
Where?
https://flashbak.com/wp-content/uploads/2016/04/Roy-Stryker-... for example is a simple fix. Its something I can do now with, even with my skills.
https://flashbak.com/wp-content/uploads/2016/04/killed-photo...
That however would require someone skilled with photoshop. You'd need to find and blend in a cat's face. Its not impossible, but its hard.
But then we get to: https://flashbak.com/wp-content/uploads/2016/04/Roy-Stryker-...
Thats not very hard to restore, but what text are you going to put in? unless you're careful its going to be misleading.
which kinda leads to the greater point, when you "restore" something, colourise or up-res, you end up putting in anachronisms, something that should be there. Perhaps its better to have these photos in context, rather than "healed" with misleading data.