NeROIC

NeROIC is a research-grade AI system for reconstructing detailed 3D object representations from unstructured, in-the-wild image collections. Given photographs captured with different cameras, lighting conditions, and backgrounds, NeROIC jointly recovers high-quality object geometry and spatially-varying material properties. The method is designed to be robust to clutter, occlusions, and inconsistent illumination, making it suitable for real-world internet photo collections rather than controlled studio captures.

At its core, NeROIC learns an object-centric representation that disentangles shape, reflectance, and lighting. This enables accurate novel-view synthesis, physically consistent relighting under arbitrary environment maps, and realistic compositing of the reconstructed object into new backgrounds. The system supports applications such as virtual product visualization, asset creation for graphics and AR/VR, visual effects, and research on inverse rendering and neural scene representations.