[nine_k]

The most interesting exhibit for me is "People watching a television set for the first time", where everything is colorized except the TV image, which correctly remains B&W. I wonder what kind of a training set provided the neural network with this notion.

[citnaj]

Author here- So I'll just be brutally honest on that one- not all renders are doing that. I cherry picked the one that did that because yeah, it's amazing. There's a simple explanation for why it sometimes doesn't pick up on the guy on tv to color it- The source material is fuzzy and small.

I wish I could claim it was something more awesome than that but that's the truth! I'm treating these outputs as an art of selection to a certain extent because it's simply not 100% consistent yet. That's one of the things I'm going to continue to try to improve upon.

[anigbrowl]

Shooting from the hip here, but I'd much rather you optimized for speed and allowed selection from a rotating palette of local maxima. I noticed your sadness about the limitations on the picture of the indian woman leaning against a tree whose hand came out red, presumably because of vignetting or some chemical inconsistency in the film substrate. But that superposition of possible interpretations on noisy data is something that shouldn't be thrown away - it's the same 'error' that makes optical illusions interesting when they seem to flip back and forth between being a vase and 2 faces or a duck vs. a rabbit. The model is doing such a great job that trying to push it too far in any one direction risks overfitting.

What I'd love to see in the future are compound networks where a few nodes like this can be mixed with a few nodes that extract vector data, a few others that infer depth maps from images, modulated by similarity detectors that match objects and individuals.

I'm very impressed by the work you've already done - I have a huge library of images I'd like to run it against for both forensic and aesthetic purposes.

[ekianjo]

> I noticed your sadness about the limitations on the picture of the indian woman leaning against a tree whose hand came out red,

I think the biggest problem is that picture is not the hand (its very visible and it could be easily fixed in post processing), it's the blue shade in the clothes that just should not be there. Otherwise, the colors are great (skin and all look very real).

[Mtinie]

Are we discounting the possibility of the NN calculations resolving to show her hand as it really existed? Tonally, it may have been differentiated from the general population in such a way the algorithm amplified this difference?

[sytelus]

If you had like to brutely honest, you should put randomly selected set, along with the hand-picked set - labeling each set how it was selected. This is a cancer in current deep learning research. You see paper with such a glowing cool examples but in reality they are just hiding all problematic cases while being fully aware of it. If this happened anywhere else in any other domain people would say they got ripped off and outright lied to.

[citnaj]

I understand the frustration and in fact share it to a certain extent with science in general. Keep in mind that this wasn't intended to be published as a paper or anything like that. I'm just a software engineer who picked a problem and found a pretty cool solution.

Primarily I thought it was cool because it should be useful in many other image modification domains. And then it blew up in popularity today (didn't expect that). But yeah in the notes in the readme at github I do say this:

>To expand on the above- Getting the best images really boils down to the art of selection.

I added that after getting some feedback similar to yours, because before that, this disclaimer wasn't quite cutting it apparently:

>You'll have to play around with the size of the image a bit to get the best result output.

So yeah I'm trying to stay honest here. I'm not going as far as picking completely random samples, admittedly, but really what I'm trying to drive at here is you can produce cool results with this tool. It's not perfect, but it's a tool. And even if you pick at random, they still look pretty damn good. Just sometimes it renders the tv as color and sometimes it doesn't, and i picked the cool option.

[citnaj]

I should add too, just to be clear- I'm not at all involved in academia.

[cmpb]

Might make a good addition to the "Known Issues" section or somewhere else. Just somewhere to indicate all the little points of human intervention.

[citnaj]

Yeah I tried pointing that out in the Known Issues section by alluding to adjusting the size parameter as a means to get the best images. But I think I'll just go ahead and be crystal clear on the "art of selection" part so that this doesn't come across as snake oil.

[citnaj]

Added this to the Known Issues section.

[sova]

to demand documentation of art is to impede the process

[citnaj]

LOL I think that'll be my go to excuse for lack of documentation for now on....

[mrharrison]

That seems like it would be quite difficult. If it is colorizing a black and white image wouldn't it colorize the black and white image on the TV screen. You would almost have to train it to recognize old TV's that produce black and white images, so that it wouldn't colorize the TV screen. Unless you can get a unique signature from a black and white photo of a black and white screen. Fun stuff.

[samstave]

Do you have test pics of a known color pic, then save a black and white version, then colorize the black and white one with this and compare?

[dang]

(I'm a moderator on HN.) Your account was being rate limited by our software. I'm sorry! We've marked it legit so this won't happen again. Please participate as much as you like.

[compsciphd]

it didn't, the tv picture is much more blue (at least to my eyes) than the original.

[kyriakos]

Even though it's blue it didn't try to colorize the man in the TV picture

[compsciphd]

couldn't that be that it can't recognize him at all?

[agitator]

AI smart enough to recognize that a dismembered head would be pale ;)

[monocasa]

The early B&W TVs were pretty blue, at least the ones I've seen. They were more monchrome than B&W.

[amelius]

Yes.

[whoisjuan]

There's some haze on the original image. I think it simples ignores the hazy portion of an image or attempts a very light colorization which seems to be the case here.

[amelius]

I'd suggest to run the code on a modified image, where the TV screen image is more clear.