OpenAI has released Point-E, which is similar to DALL-E but capable of 3D modelling. Point-E draws attention especially by being twice as fast as competing systems.
The next breakthrough that will take the world of artificial intelligence by storm could be tools that produce 3D models. Open-source Point-E, developed and made available by OpenAI, will appear in this area. In short, Point-E is a system similar to DALL-E, but differently, it transforms the desired into a 3D model.
Of course, Point-E doesn’t create 3D objects in the traditional sense because it’s just software. Instead, it creates point clouds and clusters of data points that represent the 3D shape. Just like Google’s DreamFusion tool. However, unlike Point-E, it is much faster. Systems like DreamFusion require a very long time and a powerful GPU system to render images. OpenAI Point-E requires a single Nvidia V100 card and a few minutes. The E in its name also represents efficiency. OpenAI states that it has trained an additional AI system to speed up the process.
3D modeling is used in a variety of industries and applications. Blockbuster movies, video games, CGI effects, mapping models, VR, AR and metaverse are just some of the uses. Today, text-to-image systems are quite mature. DALL-E, Stable Diffusion and Midjourney are the first ones that come to mind. However, systems from text to 3D models are in a different field and the development process continues.
Differently, Point-E combines two methods: text to image and image to 3D model. OpenAI already dominates the first stage with DALL-E and DALL-E 2. In the second phase, the AI was trained with several million data paired with 3D objects. However, this dataset is not very large, it is open to development. That’s why Point-E can sometimes create broken or missing models.
In the article shared by OpenAI, “Although our method outperforms state-of-the-art techniques in this evaluation, it produces samples in a fraction of the time. This may make it more practical for certain applications or allow for higher quality 3D objects to be discovered.” expressions were used.
With all eyes on 2D art generators right now, model synthesizing AI could be the next big industry step. However, for now, it is useful to see the issue as a starting point. On the other hand, it is necessary to be prepared for the news of artificial intelligence that produces 3D models next year. Because, as we said above, the size of this area is very large.