4 ms·

Seriously?? This is like the uber of cool or the spotify of awesome!! There's a link in this blog http://tinyclouds.org/colorize/ http://tinyclouds.org/colorize

Seriously?? This is like the uber of cool or the spotify of awesome!! There's a link in this blog http://tinyclouds.org/colorize/ http://tinyclouds.org/colorize/ where you can get the trained model for Tensorflow. I'll try this out. MIND = Blown.

StavrosK 11y ago
Seriously what the hell? How does it even know that the people in the Cartier-Bresson photo are black? I have no idea how it's doing that stuff.
- JonnieCache 11y ago
  How do you know they're black?
  - StavrosK 11y ago
    Advanced face recognition wetware.
- IanCal 11y ago
  It doesn't really need to, it just needs to expect it to be skin. I'll probably get some terms wrong here as it's been a little while since I did this kind of thing but: When you look at skin, people seem to have very different colourings. I've got freckles so I'm even patchy, and depending on the time of year will be lighter or darker. However, skin colour when you remove brightness becomes much more standard. There's remarkably little variation between a normalised colour of black and white skin, black skin is simply darker. Look at the water by the faces, it's quite a generically caucasian colour. That's because it's just lighter. Here's some old stuff I did with finger pointing detection: https://figshare.com/articles/Finger_pointing_detection/953171 https://figshare.com/articles/Finger_pointing_detection/9531... https://ndownloader.figshare.com/files/3155228 https://ndownloader.figshare.com/files/3155228 <- A plot of measured skin colours https://ndownloader.figshare.com/files/3155291 https://ndownloader.figshare.com/files/3155291
  - StavrosK 11y ago
    Ah, yes, it looks like you're right. Looks like it just used a generic "skin" colour and the luminosity in the photo made it look "right" for black skin.
  - simcop2387 11y ago
    That makes a lot of sense given that skin is colored by melanin. I would expect some small differences due to the fact that lighter skin colors are a bit more transparent but that they largely have the same base is pretty neat.

StavrosK 11y ago

Seriously what the hell? How does it even know that the people in the Cartier-Bresson photo are black? I have no idea how it's doing that stuff.

JonnieCache 11y ago
How do you know they're black?
- StavrosK 11y ago
  Advanced face recognition wetware.
IanCal 11y ago
It doesn't really need to, it just needs to expect it to be skin. I'll probably get some terms wrong here as it's been a little while since I did this kind of thing but: When you look at skin, people seem to have very different colourings. I've got freckles so I'm even patchy, and depending on the time of year will be lighter or darker. However, skin colour when you remove brightness becomes much more standard. There's remarkably little variation between a normalised colour of black and white skin, black skin is simply darker. Look at the water by the faces, it's quite a generically caucasian colour. That's because it's just lighter. Here's some old stuff I did with finger pointing detection: https://figshare.com/articles/Finger_pointing_detection/953171 https://figshare.com/articles/Finger_pointing_detection/9531... https://ndownloader.figshare.com/files/3155228 https://ndownloader.figshare.com/files/3155228 <- A plot of measured skin colours https://ndownloader.figshare.com/files/3155291 https://ndownloader.figshare.com/files/3155291
- StavrosK 11y ago
  Ah, yes, it looks like you're right. Looks like it just used a generic "skin" colour and the luminosity in the photo made it look "right" for black skin.
- simcop2387 11y ago
  That makes a lot of sense given that skin is colored by melanin. I would expect some small differences due to the fact that lighter skin colors are a bit more transparent but that they largely have the same base is pretty neat.