Apple researchers have introduced a new AI model named LiTo, capable of reconstructing 3D objects with realistic lighting effects from a single image. The model employs a unified 3D latent space representation to jointly model the object's geometric structure and light interaction characteristics, accurately reproducing complex lighting phenomena such as specular highlights and Fresnel reflections. LiTo encodes RGB-D images into latent vectors and achieves 3D reconstruction through an encoder-decoder architecture, outperforming existing models in multi-view lighting fidelity.
