DUSt3R is an AI tool that converts 2D images into 3D. It only requires a few photos to instantly extract complex data that build 3D spatial information, such as the 3D shape of an object or the camera's position and calibration.
Vincent Leroy, a researcher in AI for computer vision at NAVER LABS Europe, demonstrated how DUSt3R creates a 3D model from 2D images. Check out the video to see the amazing performance and results of DUSt3R.
Researcher Commentary
"DUSt3R is a dense multi view reconstruction tool that is super easy to use."
"Let's take two pictures of the NAVER LABS Europe offices where I work. Not changing any of the default parameters, you can very simply click on the 'Run' button, and you get a dense reconstruction in a couple of seconds."
"You can see here we also get the viewpoints from which the pictures where taken, in red and blue. It's quite robust as it can reconstruct regions that were observed by only one view like the right side of the castle."
"When we say a dense reconstruction, we mean a single 3D point per input pixel in the images. So in the end we get a complete and realistic reconstruction extremely quickly."