RenderFormer-V2: Neural Rendering with Heterogeneous Scene Primitives

Researchers introduced RenderFormer-V2, a unified transformer-based neural rendering model designed to complement physics-based rendering systems by handling diverse light-transport effects including caustics, volumetric scattering, environment lighting, textured and displaced surfaces, and out-of-core data. The model aims to improve rendering efficiency and quality for complex scenes.

We present 'RenderFormer-V2', a unified learned transformer-based neural rendering model, complementary to modern physics-based rendering systems, that can handle diverse light-transport effects such as caustics, volumetric scattering, environment lighting, textured and displaced surfaces and out-of