search

xhuangcv / humannorm

CVPR 2024: The official implementation of HumanNorm
MIT License
160 stars 6 forks source link

Fine-tuning text-to-normal Diffusion #6

Closed KhanhNgoDuy closed 3 months ago

KhanhNgoDuy commented 3 months ago

In Sec. 4.2.1, you " adapt a text-to-image diffusion model into a text-to-normal diffusion model". Can you describe this process?

xhuangcv commented 3 months ago

Hi, we fine-tuned SD1.5 using text-normal pairs. The fine-tuning code is based on the text-to-image example in Diffusers. You can find more details in our supplementary material. After fine-tuning SD1.5, we can generate normal maps by providing text prompts.