RelightFormer: Feed-forward Generative Transformer for Multiview Object Relighting

Researchers introduced RelightFormer, a feed-forward generative transformer that performs multiview object relighting by integrating multi-view cues to overcome the limitations of traditional inverse rendering and single-image generative models. The model aims to improve 3D geometry and material interaction understanding in relighting tasks.

Image relighting is traditionally tackled via complex inverse rendering pipelines, which suffer from ill-posed optimization, or single-image generative models that ignore crucial multi-view cues necessary for understanding 3D geometry and material interactions. To address these limitations, we intro