Supports text generation models and image generation models with a resolution of 512×512 and can be generated in 5 seconds.
3D content creation has made significant progress in both quality and speed. Although current feed-forward models can generate 3D objects in seconds, their resolution is limited by the intensive calculations required during training. In this article, we introduce the Large Multi-View Gaussian Model (LGM), a novel framework designed to generate high-resolution 3D models from text hints or single-view images. Our main insights are two:
(1)3D representation: Multi-view Gaussian features are proposed as an efficient and powerful representation that can then be fused together for microrendering.
(2)3D Backbone: We propose an asymmetric U-Net as a high-throughput backbone running on multi-view images that can be generated from text or single-view image input by leveraging a multi-view diffusion model. A large number of experiments have proved the high fidelity and efficiency of our method.
It is worth noting that the fast speed of generating 3D objects within 5 seconds is maintained, while the training resolution is increased to 512, thereby achieving high-resolution 3D content generation.
Note: Since the dataset used in training is based on AWS, it cannot be used directly for training in a new environment.
The necessary training code framework is provided, please check and modify the dataset implementation!
of thanks
This work is based on many amazing research work and open source projects, and thank you so much to all the authors for sharing it!
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Online experience:https://huggingface.co/spaces/ashawkey/LGM
Project address:https://me.kiui.moe/lgm/
Github：https://github.com/3DTopia/LGM
Oil tubing: