Princeton University launches Infinigen AI model to generate 3D scenes of real natural environments

IT House News on June 19, Princeton University researcher Alexander Raistrick recently released an open source AI model called “Infinigen” that can generate realistic 3D natural world scenes.

▲ Scene generated by Infinigen AI, source Alexander Raistrick

▲ Scene generated by Infinigen AI, source Alexander Raistrick

▲ Scene generated by Infinigen AI, source Alexander Raistrick

It is reported that the Infinigen model is built on the open source modeling software Blender. It can generate a series of natural "plant and animal" scenes through random mathematical rules, and can customize the terrain by inputting prompt words and add "clouds, rain, Natural phenomena such as snow, thunderstorms, and sky fires. However, the current efficiency of the Infinigen model needs to be improved. After testing by researchers, it takes 3.5 hours to generate a 1080P image (a double image of two scenes).

▲ Paper page, image source ArXiv

▲ Generated scene example, source ArXiv

In addition, the Infinigen model can also be used as a generator of training data, providing a series of functions including "object detection", "semantic segmentation", "pose estimation", "3D reconstruction", "view synthesis" and "video generation". Features can be used for scene training of computer vision-related AI.

IT House friends can find the paper address here.

▲ Generated scene example, source ArXiv

The above is the detailed content of Princeton University launches Infinigen AI model to generate 3D scenes of real natural environments. For more information, please follow other related articles on the PHP Chinese website!