Liam Hodgkinson

Sculpting NeRF Geometry: Human-Preference Fine-Tuning of a 3D-Aware Face GAN

Archer Moore, Mingming Gong, Liam Hodgkinson

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.…Read more