Request for Assets to Perform Transfer Learning Using nnU-Net

Dear Team,

Thank you for making your work open-source. The community, including myself, has greatly benefited from your contributions.

I am currently using nnU-Net and am interested in applying transfer learning to my target dataset. After reviewing the nnU-Net pretraining and fine-tuning procedure, I realized that I would need more than just the model weights to successfully perform this task.

Could you please provide the following assets to facilitate the transfer learning process?

1. Plans of the Pretraining Dataset: The nnU-Net plans file associated with the pretraining dataset. This file should include details about the network topology, patch size, batch size, and normalization schemes used during training, ensuring compatibility with my target dataset.

2. Configuration Details:

Network Configuration: Specific details about the network architecture and any modifications made to the standard nnU-Net setup.
Training Configuration: Information about the training process, including the loss function, learning rate, and any custom training schedules or augmentations applied.
Normalization Details: The specific normalization schemes used, especially if they differ from the standard nnU-Net defaults.

3. Pretraining Dataset Fingerprint (optional): If available, the fingerprint of the pretraining dataset would be helpful in understanding any dataset-specific customizations applied during training.

I appreciate your time and assistance in this matter. The provided assets will greatly support my work and the community in effectively leveraging your network weights. If I am wrong in my approach, please correct me.

Thank you once again for your contribution to the community.

Best regards, Sankarsan

Reference: https://github.com/MIC-DKFZ/nnUNet/blob/master/documentation/pretraining_and_finetuning.md

gaudot / SlicerDentalSegmentator

Request for Assets to Perform Transfer Learning Using nnU-Net #15