KidsWithTokens / MedSegDiff

Medical Image Segmentation with Diffusion Model
MIT License
979 stars 147 forks source link

Sample size training dataset #123

Closed vis6060 closed 1 year ago

vis6060 commented 1 year ago

For this particular model mentioned in the paper, what is the minimum training sample size dataset to get a prediction of the segmentation masks? Is it possible that 100 images for the training dataset, trained for 100K steps and all the parameters mentioned in the suggested python command is sufficient?

WuJunde commented 1 year ago

no, I think at least 1000 to get a fine result