Releases: axondeepseg/model_seg_unmyelinated_tem
Stanford model
result of 3rd training pass on the Stanford data
This release contains an ensemble of 5 nnunet models. This model segments myelinated and unmyelinated axon fibers, as well as nuclei and processes. Only the best checkpoints (wrt. validation score) are included.
This is a prediction example, with unmyelinated axons in green.
Full model trained on data from SickKids Foundation
These are the final results of the training performed on the SickKids Foundation data (UoT) for unmyelinated axon segmentation on TEM images. Two versions are included:
best
contains the model checkpoints that achieved the best validation score during traininglast
contains the last model checkpoints after 1000 epochs
Usually, the best checkpoints perform better, but your mileage may vary. Both versions contain 5 models from the 5-fold cross-validation scheme, for ensembling.
First nnUNet model release
This release contains a model checkpoint from one of the 5 cross-validation folds. The checkpoint was trained on data_axondeepseg_sickkids
(for more information, see the sickkids_pipeline
folder.
Please note that this model is only 1 validation fold, which does not perform as well as the 5 folds ensembled. For more information, see #1.