Rapid Interactive 3D Reconstruction from a Single Still Image

We present a method to create photorealistic 3D models from a sin- gle still image. The user provides input in the form of "scrib bles"— within a few seconds of user input, our algorithm generates a pho- torealistic 3D model and a 3D flythrough from the image. Our ap - proach uses a machine learning algorithm that learns the rel ation between (single) image features and depth. In our preliminary ex- periments, 1238 users created good 3D models for 84% of the 3775 images they uploaded. 1 Summary We consider the problem of creating a 3D model from a single still image. This problem, while interesting to both graphics and vision communities, still remains an extremely challenging one. In a nar- row mathematical sense, it is indeed impossible to recover 3D depth from a single still image, since we can never know if it is a picture of a painting (in which case the depth is flat) or if it is a pictu re of an actual 3D environment.