New AI Can Now 3D-Scan Shiny Objects Like Cars and Jewelry

A team of computer vision researchers has published a new method for turning ordinary photos into 3D models, and this one finally handles shiny things. Called RGS (Reflection-aware Gaussian Splatting), it was accepted at ICRA 2026, a major robotics and vision conference. The problem it solves is one that has quietly annoyed the whole 3D-scanning world for years.

Here's the issue. A popular technique called Gaussian Splatting builds a 3D scene from a handful of flat photos — you walk around an object, snap pictures, and software stitches them into something you can spin and zoom. It works great on matte things like wood or fabric. But shine breaks it. Reflections get mistaken for part of the object, so surfaces collapse inward, like a balloon slowly losing air. Chrome bumpers, glass bottles, and polished sneakers come out looking dented and fake.

RGS fixes this by borrowing a big pre-trained 3D AI model — think of it as an experienced sculptor who already knows roughly what shapes look like — and using it to gently correct the geometry. The team also added a check that makes different camera angles agree with each other, plus a trick that adds extra detail precisely where the shine changes. The payoff: reflective objects render smoothly instead of caving in.

Why should you care? Nearly everything desirable is shiny. Cars, watches, phones, cookware, museum pieces, diamond rings. Better 3D capture means online shoppers could rotate a photorealistic product before buying, designers could preview materials faster, and game and film studios could build convincing digital versions of real objects. The catch: this is still a research paper tested on curated objects, not something your phone does tomorrow. It needs multiple photos and real computing power. But the era of 3D scanning that chokes on anything reflective is starting to end.