OpenAI Open platform of artificial intelligence based in San Francisco has announced the issue of Point-E-System of machine learning, which allows users to generate a 3D object based on simple text input.
A team of scientists has developed a completely new approach. Point-E does not create 3D objects in a traditional sense. Instead, it creates point clouds or discreet sets of data points in space that represent a three -dimensional shape.
Generating point clouds is much easier than generating real images, but they do not capture a fine-grained shape or texture of the object-a key limit of Point-E. To bypass this restriction, the Point-E team has trained an additional AI system to convert point clouds into grid.
Point-E consists of two models: text model for image and image model up to 3D. The text model for the image, similar to generative artistic systems, such as its own Dall-E 2 OpenAI, has been trained on label paintings to understand the associations between words and visual concepts. On the other hand, the 3D image model received a set of images in combination with 3D objects to find out how to effectively explain between them.
One of the biggest advantages of this approach is that it is very fast and unjustified in terms of equipment required to produce the final image.
OpenAI researchers note that Point-E clouds can be used to produce real objects such as 3D printing. With an additional model converting the system, the system can also find a way to work flows in the field of game and animation.
“We think that the point is able to effectively produce various and complex 3D shapes conditioned on hints.
Learn more about the point in paper
The code is available Girub
