Hacker News Reader

Comment by etaioinshrdlu | Hacker News Reader

The author of the exceptional neural net (DeOldify) behind this is Jason Antic. https://twitter.com/citnaj

The funny thing is he has been kind of attacked on Twitter recently for bizarre reasons ... people are pointing out how colorization detracts from the source material, or ruins history, etc.

What this really is, is a hacker making the world's best neural net in the world at a certain task. (It appears to have also become his day job now.)

Let's try to appreciate what an achievement it really is. Pretty much all of the automatic colorization you may have seen online in the last year is some version of DeOldify.

View on HN

After watching both versions, my brain processed a different story between the two. In the colorized one, I see a bunch of people having a random good time, then some passerby gets caught in the middle. And although he tries to escape, he doesn't seem too upset. This results in a feeling of mild entertainment. Kind of like the same feeling from watching a video of household pet antics, or similar.

In the original, I see a bunch of Vaudeville "clowns" doing their antics, and a "straight-man" appears on stage, and tries desperately to escape the situation. The result is a feeling of the film attempting to induce mild amusement. Somewhat like watching a film where someone slips on a banana peel, or sits on a breakaway chair (slapstick comedy).

I believe that this is due to all the exposure I've had of the Vaudeville material has been based on grainy jittery monochrome film, so I interpret the material based on the typical Vaudeville setup.

This is true even with still photography.

I still remember an example from a series of photography books that my parents purchased for me when I was teenager that went into this.

Two scenes from an inner-city neighborhood of children playing and adults standing around. The black and white version emphasized the general grittiness. (Which by that point in the 1970s we were proabably at least somewhat conditioned to often associate B&W with.)

By contrast, the color version had lots of bright colorxs and really emphasized the children happily playing.

The video was a staged (I believe) silent film[0], so it seems like the impression you got from the original black and white version was more "accurate".

[0]https://en.wikipedia.org/wiki/Bataille_de_boules_de_neige

How much of that difference was due to the different soundtrack? I noticed the difference as well and attributed most of it to the slow and sappy music of the upscaled version versus the fast and irreverent music of the original.

Also a rather significant achievement: people are actually watching these old videos. Seems like it stands a chance of being the widest distribution of such content ever, by a large margin. When was the last time most of us consumed century-old media of any kind?

Though I understand the historian side of things... every time I've seen major colorizing or cleanup efforts, I've seen far more people looking at them and talking about the history around that time than I'd normally see in, like, a year or more. Feels like a massive net-win to me.

The difference with a colorizing, cleanup or restauration effort of a material object is that in that case the source material is not destroyed.

In that regard, the only downside for historians is that it might become more difficult to find the source material if upgraded versions of it proliferate, but this is not like rewriting or erasing history like some have claimed.

I think that as much as possible, links to the "upgraded versions" should include links to the original (the originals are public domain now in almost every case so this shouldn't be a problem). The only danger is if people don't have the originals for reference and think that details in the "enhanced" versions (that might be completely artificial) have more reality than they do; I don't think the historians have a gripe if promoting these colorized, upsampled versions is done in a way that also promotes the originals.

Classical music, Shakespeare, da Vinci, the Bible?

> I understand the historian side of things

You mean the contrarian side of things? There isn't a big movement of historians against this.

If even that. Really a journalist at Wired that put his voice behind two whole "historians." One of which, Luke McKernan, has this to say recently in the blog where quotes were lifted from:

> There has been renewed interest in this 2018 post after it was quoted in a piece by Thomas Nicholson, ‘YouTubers are upscaling the past to 4K. Historians want them to stop‘, for Wired.co.uk. The piece is chiefly concerned with the recent ‘upscaling’ of early films done by Denis Shiryaev on the Neural Love site. I rather admire this work, which is reasonably honest about how it has treated the originals, and is fun. Contrary to what the article’s title states, this historian does not want them to stop. I just want there to be better understanding, not least of the consequences.

To me it seems like there has been a much larger issue of photographs, movies etc. being digitalised at some point years ago using some standard of the time, after which the originals were discarded because "there, now it's saved and the old ones are an inferior media anyway".

Then you're left 20 years later with a horribly jpeg'd mess in 800x600 px.

I don't have a source on this, but it's my impression a lot of libraries did this in the 90's or 2000's (if someone knows specific cases i'd love to hear about them, I hope I'm wrong)

Reply on news.ycombinator.com