Generating colorized digital images utilizing a re-colorization neural network with local hints

A system comprising:

• one or more memory devices comprising a two-stage image colorization neural network, the two-stage image colorization neural network comprises an image colorization neural network and an image re-colorization neural network;
• and
• one or more processors configured to cause the system to:
• generate, utilizing the image colorization neural network, a first modified digital image from a digital image by colorizing the digital image based on a luminance channel of the digital image, the first modified digital image being a colorized version of the digital image;
• receive, via a graphical user interface displaying the first modified digital image including a plurality of predicted pixel values colorizing a plurality of grayscale pixels of the digital image, one or more user inputs to one or more selected locations with predicted pixel values within one or more regions of the first modified digital image;
• determine, from the one or more user inputs for the one or more selected locations within the one or more regions of the first modified digital image, one or more local hints indicating one or more selected colors to modify color values of the one or more selected locations of the first modified digital image;
• and
• generate a second modified digital image from the first modified digital image utilizing the image re-colorization neural network to process the one or more local hints, a luminance channel of the first modified digital image, and color channels of the first modified digital image by modifying one or more colors corresponding to the one or more regions of the first modified digital image by changing predicted pixel values generated by the image colorization neural network in the color channels of the one or more regions according to the one or more selected colors indicated by the one or more local hints.

A computer-implemented method comprising:

• generating, utilizing an image colorization neural network, a color digital image comprising chrominance values from a grayscale digital image by generating a plurality of predicted pixel values colorizing pixels of the grayscale digital image, the color digital image being a colorized version of the grayscale digital image;
• determining, by at least one processor, a luminance channel and color channels corresponding to a plurality of pixels of the color digital image;
• receiving, via a graphical user interface displaying the color digital image including the plurality of predicted pixel values colorizing the pixels of the grayscale digital image, one or more user inputs to one or more selected locations with predicted pixel values within one or more regions of the color digital image;
• determining, from the one or more user inputs for the one or more selected locations within the one or more regions of the color digital image, one or more local hints indicating one or more selected colors to modify color values of the one or more selected locations of the color digital image;
• and
• generating, by the at least one processor utilizing an image re-colorization neural network, a modified digital image from the color digital image by processing the one or more local hints, the luminance channel, and the color channels by modifying one or more colors corresponding to the one or more regions of the color digital image by changing predicted pixel values generated by the image colorization neural network in the color channels of the one or more regions according to the one or more selected colors indicated by the one or more local hints.