Yoshiharu Sato, creator of Gauzilla the 3D Gaussian Splatting renderer written in Rust, has been steadily adding features to Gauzilla Pro.
Most recently Sato had introduced the ability to provide annotations within a capture through a vision transformer (ViT) based segmentation model, which allows for completely serverless operations and GPU independent inferencing.
Now you're able to move, rotate, clone, or erase objects interactively. In addition, you can create segments and labels on a per object basis. For instance, if you wanted to change the location of a sofa, it is labeled within the environment. If you're wondering what happens when you remove a section of a scene, it appears that he's using some sort of inpainting to patch it up.
As Yoshiharu demonstrates, it's easy to see this being applied to interior design and product visualizations.
This is a welcome complement to the existing feature set. Gauzilla Pro is able to be run on the web without any user installation, making this an easy tool for businesses to plug and play.
Currently Gauzilla Pro is only accepting waitlist applications. The standard version of Gauzilla remains MIT Licensed, inviting further exploration and development. Find Gauzilla on its GitHub page and follow Sato on LinkedIn for updates to the project.
