InnerEye-DeepLearning
InnerEye-DeepLearning copied to clipboard
microsoft
Medical Imaging Deep Learning library to train and deploy 3D segmentation models on Azure Machine Learning
[pytorch-lightning < 1.5.8 is vulnerable to Deserialization of Untrusted Data](https://github.com/advisories/GHSA-2vj5-px25-gjrp). We are using 1.5.5: https://github.com/microsoft/InnerEye-DeepLearning/blob/914a89383d3275cb3304bde9c6cdd6050357f3f7/environment.yml#L57 Should we use a newer version? [AB#4978](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4978)
[AB#4816](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4816)
On RadiomicsNN: Run model "--model=PANDA_SimCLR", while setting local_dataset of PANDA_SimCLR to None. This will bring an error `No such file or directory: '/tmp/tmp3pop2156/PANDA_tiles_20210926-135446/panda_tiles_level1_224/dataset.csv'` dataset is mounted, but not at the...
HD_c76edbb2-7d26-445b-9571-90c5c91cd010_0 in RadiomicsNN, experiment antonsc_recovery2, pl_1_5_crossval. This job fails with a service error in hosttools. Root cause is that we are deleting large sets of files after finishing the crossvalidation...
[AB#4745](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4745)
`create_transforms_from_config` hardcodes what can be used, and is not extendable [AB#4743](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4743)
Somewhere along the process from data loaders to training_step we seem to lose metadata. Lost in the to_device methods? [AB#4742](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4742)
[AB#4716](https://innereye.visualstudio.com/60ce1777-00d6-4015-82bc-488a0c00202f/_workitems/edit/4716)
The method `MLRunner.run_inference_for_lightning_models` takes a list of checkpoint paths as an argument, but then makes sure that there is only one used ([here](https://github.com/microsoft/InnerEye-DeepLearning/blob/3fc71b847e9dd4ad9d0b070722558064647d95a2/InnerEye/ML/run_ml.py#L444-L445)): ``` if len(checkpoint_paths) != 1: raise ValueError(f"This...
I thought that this had been fixed in PL, but it seems not. This job in RadiomicsNN raises an exception in one of the child processes, but the main process...