The website for this project is here. They used a CNN trained on 1 million black & white/color paired images from the Imagenet dataset. In a human experiment they were able to fool participants on the generated image versus the ground truth color image in 32% of the trials.
tristanburke
Not so much a computer generated coloring as Andrew linked above. But this reminded me of the recent film "They Shall Not Grow Old" directed by Peter Jackson. This film restored WW1 footage with added color. https://www.youtube.com/watch?v=_cSXfKSRKz4
hershg
A bit different but kind of interesting, there's black-white images from before the color-camera invention, where the photographer put a red, green, or blue film in front of the camera - thus creating three b-w images representing the r, g, and b channels. Today we can reconstruct the actual images by interpolating these three images. (Note this is from something i saw in cs194 comp. photography)
The website for this project is here. They used a CNN trained on 1 million black & white/color paired images from the Imagenet dataset. In a human experiment they were able to fool participants on the generated image versus the ground truth color image in 32% of the trials.
Not so much a computer generated coloring as Andrew linked above. But this reminded me of the recent film "They Shall Not Grow Old" directed by Peter Jackson. This film restored WW1 footage with added color. https://www.youtube.com/watch?v=_cSXfKSRKz4
A bit different but kind of interesting, there's black-white images from before the color-camera invention, where the photographer put a red, green, or blue film in front of the camera - thus creating three b-w images representing the r, g, and b channels. Today we can reconstruct the actual images by interpolating these three images. (Note this is from something i saw in cs194 comp. photography)