Sculpting NeRF Geometry: Human-Preference Fine-Tuning of a 3D-Aware Face GAN (opens in new tab)

Reinforcement learning from human feedback (RLHF) for 3D generation is now established across a number of works, but most existing pipelines optimise explicit surface representations, often by converting radiance fields into meshes and training heavily on surface-supervised data. We instead fine-tune a pretrained 3D-aware generative model directly from a learned reward over radiance-field density ($\sigma$) values, with no externally supplied me...