New Disney Research Algorithm Capable of Turning 2D Photos into 3D Models

Disney Research developed an application that can build 3D models from ordinary photos. Thanks to the original algorithm, the program is capable of building several hundred images of three-dimensional models of complex scenes straight from real life, and therefore promises to be a useful tool for creating films, games and models for 3D printers.

Building a 3D model from several two-dimensional images that captured an object from different angles is no longer a novelty, but creating high-detailed environments in high resolution has proved to be a difficult task due to the need to process a large amount of data. The merit of Disney Research is that the algorithm they created is able to effectively manage such data scales, processing them without having to keep all the data in memory at the same time.

Now many 3D models are obtained by laser scanning. In the case of complex environments, one laser may simply not be able to cope with the number of parts due to the fact that some objects will obscure others. The method of processing ordinary photographs makes it easier to “capture” a scene from several angles, but building 3D models from them until recently has been very difficult.

The authors of the algorithm demonstrated its work on complex scenes of external and internal spaces, which were recreated in 3D based on images taken by a 21-megapixel SLR digital camera. For comparison, before that, all reconstruction techniques were able to process only images with a resolution of 1-2 megapixels. To create each of the models, 100 shots were required.

The developers claim that the new approach can be applied to photographs taken with conventional digital cameras.



The study is published in the report " Scene Reconstruction from High Spatio-Angular Resolution Light Fields ".

Source: www.3ders.org