gobunu/HAODiffPython
(20 stars) (1 fork) (0 issues indexées) (0 good first issues ouvertes)
Dépôts
We stably expand the latent channels of a pretrained diffusion restorer, enabling efficient one-step human restoration with higher fidelity and fewer artifacts.
We propose a physically grounded one-step diffusion method for face fill-light enhancement, trained on our 160K-pair dataset with 6D area-disk lighting control and physics-aware conditioning.