yandex-research / ddpm-segmentation

Label-Efficient Semantic Segmentation with Diffusion Models (ICLR'2022)
https://yandex-research.github.io/ddpm-segmentation/
MIT License
662 stars 60 forks source link

the best diffusion steps and UNet blocks #27

Closed Ghy1209 closed 1 year ago

Ghy1209 commented 1 year ago

Hi, thanks for your working. Can you share the best diffusion steps and UNet blocks using for each task of of DDPM-based segmentation? By the way, can i use the step 0 ?

dbaranchuk commented 1 year ago

the steps and blocks for each dataset are presented in the corresponding config files. These steps are not perfectly tuned but should work fine.

can i use the step 0

According to our experiments, the segmentation quality gets worse but it still provides reasonable performance.