Nvidia presented research this week at CVPR 2022 in New Orleans on a method for turning a series of photos into 3D objects with the help of increased compute power.
Called Nvidia 3D MoMa, the company believes it will one day help designers import an object into a graphics engine. There, they can modify the object’s size or change it in different ways, including the materials or different lighting effects. Ultimately, the research hopes to reduce the time required for content creators to produce graphics.
The method uses inverse rendering, a widely used technique to reconstruct a series of photos into a 3D model. In this case, Nvidia uses GPU acceleration and AI to quickly produce 3D objects, said David Luebke, vice president of graphics research in a blog on Tuesday. https://blogs.nvidia.com/blog/2022/06/21/inverse-rendering-3d-research-cvpr/
Nvidia 3D MoMA generates triangle mesh models of an object within an hour using Nvidia Tensor Core GPUs.
Researchers worked with creative teams at Nvidia to collect 100 images of each of five jazz band instruments from different angles. Then 3D MoMa reconstructed those images into 3D mesh representations. Next, they used Nvidia Omniverse for edits to dress the mesh to create materials for the instruments such as gold or wood. They became the building blocks for an animated scene shown in a video.
The research is described in a paper Nvidia shared at the New Orleans conference on Computer Vision and Pattern Recognition. Nvidia shared a short video of the process.
Related: Oppo demonstrates ray tracing 3D wallpaper for smartphones