The Klimt Color Enigma

Awesome work! Btw, you could have also submitted this as Show HN :).

Amazing work! Congratulations!

Is there anywhere the actual restored artworks in the article can be seen without having to navigate a lot of Web 3.0 gimmickry, or without having to scale up some 2000-era images the size of postage stamps?

What is the difference between coloring and colorizing?

I'm a big Klimt fan and am very impressed with this (especially Jurisprudence). I was thinking, though: "this is so hard to see on mobile. Maybe better on desktop."

Speaking only for myself, no! It’s just borrowing and playing with ideas, which artists do all the time. (I have an education in fine arts, for whatever it’s worth [0])

[0]: nothing, both literally and figuratively

They have human Klimt experts helping too.

Klimt is dead, and the originals are missing.

Not exactly made up. Having only the lightness values to start with, The task is to define the hue and saturation. Though in theory there is no way to know for certain what they originally were, in practice certain lightness values are associated with some hues more than others. Also, saturation values tend to follow lightness values quite closely. Add to that the fact that Klimt (like most painters) employed a habitual pallet, you are half was to a good guess.

I empathize with this position, but I think this situation is closer to authentic restoration (which the art world is okay with, generally speaking) than an unguided machine transformation of a human work.

(I've visited just about every Klimt exhibit in Vienna, including the one that these B&W photographs are in. I think they've done a remarkably faithful job here, down to accurately capturing the subperiod of Klimt's "gold period").

It seems different when we're just trying to guess what the actual colors were, since the artist is dead and the painting is destroyed, and we only have a black and white copy + textual description of the painting.

No, absolutely not.

I think that one has to be careful with labeling - these works are highly researched guesses based on lots of sources but are not "the article itself." But I think this process, where they are trying to be a thorough as possible in guessing what colors Klimt might have used mechanistically, is much more respectful than other techniques. Consider the recent Vermeer restoration[1] (hn discussion at [2]): 'restoring' the painting in this way alters the actual article and could be altering the intent of the artist. I support the work, but I think there's a basic element of uncertainty to it.

At the end of the day, if you don't want to accept the machine's guess at the colors Klimt used, you don't need to! You have access to the same archival documentation of what his lost paintings looked like as everyone else.

[1] https://hyperallergic.com/672345/vermeer-restoration-finally...

[2] https://news.ycombinator.com/item?id=28468802

Unfortunately not, but here are a few great public art datasets: https://www.artnome.com/art-data. Rijksmuseum and wikiart also have available datasets.