Ell-M

Integration of Pretraining and Fine-Tuning Models into Refactored GDL

1

A project enabling self-supervised pretraining on satellite and aerial imagery was previously developed in a separate repository. This functionality must now be fully integrated into the main branch of the...

[BUG] Incorrect mIoU computation in test phase

1

Issue in the test evaluation stage where the current manual checkpoint loading mechanism in [geo_deep_learning/train.py](https://github.com/NRCan/geo-deep-learning/blob/main/geo_deep_learning/train.py#L51) does not correctly restore the best model. Replace the existing logic with the following PyTorch...