OpenAI launches Shap・E: Generate high-quality 3D models based on text in seconds

News on May 9th, following the text generation image model DALL・E, OpenAI recently released the Shap・E model again. Users can enter text that is used to create realistic and diverse 3D models.

Shap・E is not just a 3D model generator, but can directly generate parameters of implicit functions that can render texture meshes (textured meshes) and neural radiation fields (NeRF).

This means that Shap·E, unlike current models that only output point clouds or voxels, can generate high-quality 3D assets with fine-grained textures and complex shapes.

OpenAI shows the results of Shap・E, such as a bowl of food, a penguin, a voxelized dog, a campfire, an avocado-shaped chair, etc. The entire image can be rendered in seconds.

Attached explanations of related terms are as follows:

Neural Radiance Field (NeRF):

Neural Radiance Field (Neural Radiance Field, NeRF) As a new field of view synthesis technology with implicit scene representation, it has attracted widespread attention in the field of computer vision.

As a novel view synthesis and three-dimensional reconstruction method, the NeRF model has been widely used in robotics, city maps, autonomous navigation, virtual reality/augmented reality and other fields.

Point clouds:

Cloud points are a series of points falling in a three-dimensional coordinate system. They can be expanded on the building surface or building facade to capture the appearance of the building. form.

Voxels

Voxels are pixels in 3D space. Quantitative, fixed-size point clouds. Each cell has a fixed size and discrete coordinates.

The above is the detailed content of OpenAI launches Shap・E: Generate high-quality 3D models based on text in seconds. For more information, please follow other related articles on the PHP Chinese website!