Jun 20, 2024 16:00:00

I tried out 'Unique3D', which can generate high-quality 3D meshes from a single image at high speed.

A research team from Tsinghua University in China has released a framework called ' Unique3D ' that automatically generates a 3D mesh model from a single photo. Unique3D predicts the shape of a subject from a photo taken from one direction by applying a 2D diffusion model, and can generate a 3D mesh model relatively quickly. A free online demo is also available.

[2405.20343] Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image

https://arxiv.org/abs/2405.20343

GitHub - AiuniAI/Unique3D: Official implementation of Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image
https://github.com/AiuniAI/Unique3D

At the time of writing, the online demo of Unique3D was available on the online AI platforms Hugging Face and Gradio . This time, I actually tried using the Unique3D demo with Hugging Face.

Unique3D - a Hugging Face Space by Wuvin
https://huggingface.co/spaces/Wuvin/Unique3D

When you access the demo page at the URL above, it will look like this. Click on the area that says 'Drop your image here -or- Click to upload' to upload your image.

The image I uploaded this time is a photo of

iCEO that I quickly took with my smartphone.

Once you have uploaded your image, click on 'Generate 3D' on the right.

Then, a 3D mesh model was generated in about 60 seconds.

The three sides of the 3D mesh model are lined up: the front, left side, and back. The original was only one photo taken from the front, but the sides and back were also properly modeled. However, the generated 3D mesh model appears to be a little flatter than the real thing.

The generated 3D mesh model can be saved locally by clicking the download icon in the upper right. The file format is glTF format.

According to the Unique3D development team, Unique3D uses code from Stable Diffusion , Wonder3d , Zero123Plus , Continues Remeshing , and Depth from Normals .