Can you tell me if the UNet and VAE blocks in stable diffusion used in the model use the official pre-trained weights provided by stable diffusion or your own trained weights?

YuzheZhang-1999 / DiffTSR

[CVPR2024] Diffusion-based Blind Text Image Super-Resolution (Official)

63 stars 4 forks source link

Can you tell me if the UNet and VAE blocks in stable diffusion used in the model use the official pre-trained weights provided by stable diffusion or your own trained weights? #1

Closed zuiaichirouya closed 3 months ago

zuiaichirouya commented 3 months ago

I'm sorry to bother you. Can you tell me if the UNet and VAE blocks in stable diffusion used in the model use the official pre-trained weights provided by stable diffusion or your own trained weights?

YuzheZhang-1999 commented 3 months ago

All modules are trained from scratch without any pre-training weights.

zuiaichirouya commented 3 months ago

Can I ask again how much computational resources would be required to train this model?

YuzheZhang-1999 commented 3 months ago

Each module is trained on a machine equipped with 8 NVIDIA A100-40GB GPUs.

zuiaichirouya commented 3 months ago

Thank you very much for your answer.