Interesting work titled "Shap-E: Generating Conditional 3D Implicit Functions" from OpenAI. Shap-E is a generative model for creating 3D objects.
The model is trained in two stages: first, the encoder is trained, and then the diffusion model is trained on the outputs of the encoder. Training on a large dataset with 3D and text data allows for the creation of complex and diverse 3D objects in seconds. Compared to Point-E, an explicit generative model based on point clouds, Shap-E converges faster and achieves comparable or better quality.
Paper: https://arxiv.org/abs/2305.02463
Github: https://github.com/openai/shap-e
Demo: https://huggingface.co/spaces/hysts/Shap-E