Spaces:
Running
on
Zero
Apply for community grant: Academic project (gpu)
Hello!
This project aims to extend DUSt3R and MASt3R to predict interactive 3D Gaussian Splats from one or two images, without using any camera extrinsics, intrinsics or depth information. This Hugging Face page will serve as a demo to accompany the paper 'Splatt3R: Zero-shot Gaussian Splatting from Uncalibarated Image Pairs', which will shortly be published on ArXiv. The paper is a collaboration between the Active Vision Lab and the Visual Geometry Group at the University of Oxford.
The demo accepts one or two images from the user, runs it through a feed-forward transformer model to predict one Gaussian for each pixel in each image, and then displays the resultant 3D Gaussian Splat. There is also a gallery of pre-generated splats that the user can explore. We do not require persistent storage, but our demo would be greatly quickened by having the model run on GPU rather than on CPU.
Thanks for your time!
Hi @brandonsmart , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of this page so your Space can run on ZeroGPU.