Learn about Google's newest AI, LOLNeRF. Our ability to understand 3D shapes from the 2D images we see is an important aspect of human vision. On the ground, gaining such an understanding of computer-based vision systems was a major challenge. Many successful approaches rely on multi-viewdata, which makes it much easier to infer the 3D shape of objects in images by providing two or more images of the same scene from different perspectives.
However, there are many situations in which knowing the 3D structure from a single image would be useful, but this problem is usually difficult or impossible to solve. For example, it may be difficult to distinguish between an image of a real beach and an image of a flat poster of the same beach. However, 3D structures can be estimated based on what types of 3D objects occur frequently and what similar structures look like from different perspectives.
In "LOLNeRF: Learn from One Look", presented at CVPR 2022, Google scientists provide a framework to learn how to model the 3D structure and appearance of single-view image collections. LOLNeRF learns the typical 3D structure of a class of objects, such as cars, human faces or cats, but only from single views of any one object, never the same object twice.
Click on the link below to read it more detail: