tag:blogger.com,1999:blog-4912713243046142041.post8171190123873243851..comments2024-03-28T19:45:08.560-05:00Comments on TYWKIWDBI ("Tai-Wiki-Widbee"): Modernized videos from 1911-1913Minnesotastanhttp://www.blogger.com/profile/01382888179579245181noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-4912713243046142041.post-63755083049265376652020-07-08T16:51:55.479-05:002020-07-08T16:51:55.479-05:00Yes these are fascinating. The original videos in ...Yes these are fascinating. The original videos in quality/size are shown quite early around t=9s. Both the "upsampling" (i.e. adding information to blow up the image size without pixelation) and the colorization are being handled by a machine learning algorithm. The algorithm has been trained on some higher quality film or photos and it is trying to make the historical film more like the film it was trained on. So the colors will be dependent on that training set. I'm sure if you trained it on a film like Pleasantville (1998) you could get very strange output.nolanddahttps://www.blogger.com/profile/12708768497461983779noreply@blogger.comtag:blogger.com,1999:blog-4912713243046142041.post-90281250827694768172020-07-08T12:49:07.449-05:002020-07-08T12:49:07.449-05:00I love these videos! My favorite, from the same c...I love these videos! My favorite, from the same channel, is pre-earthquake footage of San Francisco:<br /><br />https://www.youtube.com/watch?v=VO_1AdYRGW8IronHorsehttps://www.blogger.com/profile/01106255311347964531noreply@blogger.com