Using machine learning to de-age old photos of downtown Norfolk and other places of note and bring them to life. Starting with the original photos from the Norfolk Public Library, the photos are first run through Deoldify for colorization, then Real ESRGAN for upscaling.

The resulting colored, upscaled photos are then run through an “inpainting” process created by Virginia Tech’s Vision and Learning Lab. This process approximates the depth in an RGB image which in turn can be used to simulate 3D camera movement by “filling in” missing perspective in the image.

Norfolk has always been a busy, lively city and this was a fun way to convey that with experimentation.