OpenAI YesElon Musk Creating artificial intelligence that supports popular SLAB Text to image generator. recently,OpenAIAnnouncing the release of its latest image creation tool POINT-Ecan be generated directly from text prompts 3DPoint cloud. It is understood that existing systems (e.g. Google of DreamFusion) usually takes several hours and GPU to generate images, and Point-E you only need one GPU And it’s done in a minute or two.
Open artificial intelligenceable
3DModeling has been used in the development of various industries, from car parts, cars, airplanes and architectural design to modern blockbusters, video games,VR And RA of Synthetic image The effect is even moreNASA lunar crater mapping mission,Google heritage conservation projects and Meta of Metaverse vision depends on 3DModeling capabilities.
However, creating3DImaging remains a resource- and time-intensive process, 3D text is a branch of this research. Unlike similar systems,Point-EUses large corpora (text, image), allowing it to follow diverse and complex prompts, in order to generate textual prompts3Dobject,OpenAI The image is first sampled using a text-image model, then the3DThe object is sampled. Both steps can be completed in seconds and do not require expensive optimization procedures.
If you enter a text prompt, such as“A cat eats a burrito”,Point-E A composite view of the burrito-eating cat will be generated first3Dgive back. The resulting image is then passed through a series of diffusion models to create a representation of the initial image. 3D、RGB Scatterplot, generate a draft first 1,024 point cloud model, then generate a more detailed model 4,096 Scatter cloud model. In practice,Point-EIt is assumed that the image contains relevant information from the text and that the point cloud over the text is not explicitly constrained.
These diffusion models are respectively“Millions” 3D Models are trained and all models are converted to a standardized format. at presentOpenAI Already in GitHub The open source code for the project was released on .
Artificial intelligence and industrial design
Although currentlyOpenAI publishedPoint-EThe target problem is converting text to three-dimensional models, and it does not involve deep modeling like industrial design. But in3DAccording to Science Valley,OpenAI The business model to be disrupted is to sell software as a serviceSaaSModel changes to model, i.e. salesMaaSmode, then the next stepPoint-EArtificial intelligence modeling entering the field of industrial design could be just around the corner.
In the book “How is artificial intelligence changing industrial design?” » written by Zhang Xiaojun of Anshi Asia Pacific. “The article states that compared to the digital intelligence of manufacturing and business processes, the development of intelligence and automation in design is relatively slow. In many areas, the adoption ofGUJATThe inefficient design process based on drawings has even become a bottleneck for the company’s overall product development or capacity expansion, and the large number of designers will also bring significant costs to the company.
According to “How will artificial intelligence change industrial design?” » 》 article, design driven by intelligent algorithms is completely dynamic, whether it is movement that responds to the constraints of geometric relationships, or movement or deformation of objects in the physical world real under the action of force, this can be achieved in the design process. . Geometric models in any state can be exported. In this dynamic and intelligent design process, simulation has become a real tool to drive design, and forces can be used directly to shape shapes, just like nature.
Source: 3D Science Valley
Daguang focuses on providing solutions such as precision CNC machining services (3-axis, 4-axis, 5-axis machining), CNC milling, 3D printing and rapid prototyping services.
