Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Colorizing and restoring old images with deep learning

github.com

Post reply on HN
Print view View original

11–20 of 107 posts

Re: Colorizing and restoring old images with deep learning

#11

Post by drcode »

jbattle wrote: post #6

This seems almost too good to be true. One thing I find very striking is how it gets skin tones very plausible across people of different ethnicities (though the majority of subjects in the picture appear of european descent). Unless a) my brain is applying more interpretation to these pictures than I realize or b) the author (intentionally or not) picked out pictures that show the best results

Yeah some of the details are absurdly good, especially the picture of the "Texas Woman", how it gets the dogs ears perfect, perfect colors on the apples, and renders the copper pot a perfect copper hue.

Re: Colorizing and restoring old images with deep learning

#12

Post by shawn »

jbattle wrote: post #6

This seems almost too good to be true. One thing I find very striking is how it gets skin tones very plausible across people of different ethnicities (though the majority of subjects in the picture appear of european descent). Unless a) my brain is applying more interpretation to these pictures than I realize or b) the author (intentionally or not) picked out pictures that show the best results

Your brain is applying far more interpretation to color than people are consciously aware of.

https://raw.githubusercontent.com/jantic/DeOldify/master/res...

His face is arguably too red. But on average it’s fine. (Amusing: is this comment correct, or unconsciously biased by the lack of knowledge of what native Americans actually look like? I admit the latter is possible.)

Humans interpret colors thanks to context. When you strip away context, it’s easy to come up with things that fool you. (Optical illusions are the limit case of this.)

See DaVinci’s journals on color. They are worth studying, and most entries are so short they may as well be tweets. http://www.sacred-texts.com/aor/dv/dvs005.htm

Re: Colorizing and restoring old images with deep learning

#13

Post by user-x »

It's interesting to see how the algorithm seems to turn aerial photos into romantic paintings. My guess is that the model was trained on mostly up-close photos and that the colors don't map exactly to aerial photos because color intensity fades over large distances.

On that note it is cool to see how the algorithm does work for both indoor and outdoor photos. Indoor settings tend to have dark backgrounds and outdoor settings have light backgrounds.

Very cool project.

Re: Colorizing and restoring old images with deep learning

#14

Post by misnome »

dsfyu404ed wrote: post #4

Is it just me or do the example images seem overly biased toward coloring clothing as blue?

It's listed under known issues:

> The model loves blue clothing. Not quite sure what the answer is yet, but I'll be on the lookout for a solution!

Re: Colorizing and restoring old images with deep learning

#15

Post by crazygringo »

Very interesting... seems to basically learn that:

Faces -> some variety of flesh-colored from light to dark

Fabric/clothing -> blue

Sky -> blue

Vegetation -> green

Wood -> brown

Blank -> turquoise or tan

Small details -> fascinating variety of colors, but often a brilliant red

Which all seems fairly reasonable. For many things (like wood or skin) it seems accurate.

Obviously things like clothes come in such a variety of colors that there's simply no way at all to predict accurately, zero meaningful signal -- so if it settles on whatever the most common color is, it doesn't surprise me that would be blue.

Re: Colorizing and restoring old images with deep learning

#16

Post by compsciphd »

nine_k wrote: post #2

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.

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

Re: Colorizing and restoring old images with deep learning

#17

Post by kyriakos »

compsciphd wrote: post #16
nine_k wrote: post #2

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.

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

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

Re: Colorizing and restoring old images with deep learning

#18

Post by monocasa »

compsciphd wrote: post #16
nine_k wrote: post #2

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.

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

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

Re: Colorizing and restoring old images with deep learning

#20

Post by michrassena »

This is one of the few colorizing algorithms that I've seen which creates desirable output. The images really do look like old colorized images. I wonder how the authors dealt with the differences in spectral sensitivity of their source material. There's clearly some orthochromatic plates or film being used. The image of the Seneca native 1908 is a good example. Notice how dark the field is on the patch on her skirt. With orthochromatic emulsions, the patch could have been either black or red since the emulsion isn't sensitive to red. It's most sensitive to blue, which is part of the reason skies look so white in old photos.
Post reply on HN