Researchers used a neural network to reconstruct a lost Picasso painting

That has... so many obvious artificial flaws it’s not even funny.

I’m surprised I got a downvote. Compare the fake (e) to the real (d). The algorithm appears to have gotten the broad color scheme correct, but that’s about it as far as the similarities go. There are uncomfortable smudges all over the painting, the lighting makes no sense, and several details have blurred together around the hands, feet, everything.

> They have taken a manually edited version of the x-ray images of the ghostly woman beneath The Old Guitarist and passed it through a neural style transfer network. This network was trained to convert images into the style of another artwork from Picasso’s Blue Period.

Perhaps a more accurate title would be something like "Researchers used a neural network to stylize a manually retrieved lost Picasso painting"

Original paper: https://arxiv.org/pdf/1909.05677.pdf

This looks like poop. It didn’t even do a good job of faking brushstrokes, much less guessing at what the colors might have been. It’s a half-assed job that’s not doing much more than a style transfer phone app could do.

Doing this right, imho, would entail taking x-rays of every Blue Period Picasso possible, and training a network to output the visible image when given an x-ray. I don’t know offhand how many paintings were in that period, so there may not be enough work to usefully do this.

This is a neat idea, but the execution was kind of minimum-viable-attempt.

You can do this yourself though, just google some famous paintings with X-RAY images and try it. I doubt this is very accurate but it is fun.

I posted a few https://twitter.com/jonathanfly/status/1175966107091574784

This title is misleading. The painting was recreated, not retrieved. The title of the article itself is "This Picasso painting had never been seen before. Until a neural network painted it."

So much "research" is just people doing normal mundane work but they happen to be familiar with the process of publishing research papers so they do that for whatever they can think of. This particular work is more like a hobby or perhaps some curiosity for art historians to look at. They didn't do anything that we couldn't already and obviously do anyway.

Jesus it looks like crap.

Compare with the two other painting, the guitarist and the one they used for style transfer, it just has nothing to do.

Maybe the problem is that they took too much of the underlying painting, which was either very unfinished or damaged.

Anyways, this is quite useless.

"Computers are useless they can only give you answers" - Picasso

Top third of the screen is covered with a cookie warning, bottom third is warning me that I’ve read 1 of 3 free articles, when I scroll down to read this article they already count as read, there’s a big empty space, followed by a prompt to subscribe to some newsletter, followed by four lines of said article, which fade over into a button to actually read it - followed by an attempt to distract me with a different article.

I get the hint, you really don’t want me to read the article, so I won’t.