Editing Images in 3D

Often, we get asked to find ways to animate and dimensionalise images. This is usually for historical documentaries that want to bring archive imagery to life. Well, researchers at Jagiellonian University and University of Cambridge have devised MiraGe, their method for editing and animating 2D images by turning, making them 3D.

The process takes an element within an image and interprets it in 3D using Gaussian Splatting (check out some of our explorations into GS in previous posts), this allows the user to manipulate this element in 3D, then the process applies the edited area back onto the 2D image. What I love about this process is how crazy it sounds to turn an image into a 3D representation so you can do things like simulate physics, and then turn it back into an image.

The other interesting thing about this approach is that, unlike the process we used for last's week experiment, this process doesn't use any generative AI (although I think some of their examples use generated 2D images as a starting point).