Hi Organization team, Could you please clarify if the use of a generative model for data augmentation is permissible under the challenge rules, which state, "No additional kidney pathology data is allowed"? Thank you, Chloe

Created by Chloe Tran chloetran
Hi @huoy1 , thanks for your reply. That's great information.
Yes, that is allowed! Thanks for your clearance. If the pre-trained model or generative model did not use kidney pathology data, then you can fine-tune that model for this challenge.
Hi @huoy1, Thanks for your prompt response. I'm sorry for not clarify my question. I would like to ask could we fine-tune pre-trained model/generative models (which is not pre-trained on any kidney dataset) on the kidney dataset **provided by the challenge**?
If the generative model was trained by any kidney pathology data, the answer is **NO**. However, if the generative model was trained by other data types (e.g., natural images, radiology data, pathology data but not from kidney), the answer is yes. We just want to evaluate the segmentation performance of different models when they only see kidney pathology data provided by this challenge. Similarly, the same rule applies to data augmentation for pre-trained models. If the augmentation or pre-trained model does not utilize kidney pathology data (e.g., ImageNet), then you are permitted to use it in your training. We will be updating this rule on our website.

Inquiry: Use of Generative Model for Data Augmentation page is loading…