Robust 3D-Masked Part-level Editing in 3D Gaussian Splatting with Regularized Score Distillation Sampling

Summary

This paper talks about RoMaP, a new system that allows very precise editing of 3D objects created with a technique called 3D Gaussian Splatting by using special masks and a method to improve editing accuracy.

What's the problem?

The problem is that editing specific parts of 3D objects in Gaussian splatting is hard because existing methods don’t handle detailed, local changes well, making it difficult to accurately edit small or complex areas.

What's the solution?

The authors developed RoMaP which creates strong 3D masks that cover just the parts of the object to be edited and uses a special training method called regularized Score Distillation Sampling to guide the editing process. This combination helps in making detailed and accurate edits to the 3D models without messing up other parts.

Why it matters?

This matters because it lets artists and developers edit 3D models more precisely and easily, improving applications in gaming, animation, virtual reality, and other fields where detailed 3D editing is important.