HoVerNet training pipeline

[KumoLiu]

Fixes Project-MONAI/MONAI#4878

Description

HoVerNet training tutorial

• ignite version pipeline

Checks

• Notebook runs automatically ./runner [-p <regex_pattern>]

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[bhashemian]

Hi @KumoLiu, since we only have ignite based pipelines here, could you please update the name of the files to evaluation.py and training.py? thanks

[JHancox]

Apart from the CV2 comment I made, it looks good.

[OmarAshkar]

Hi @KumoLiu Thanks for your great work. I have been looking for the best hovernet implementation for my case. However, I have a comment here on using .mat format.https://github.com/KumoLiu/tutorials/blob/1b9f64d3a32fa08d37048dab282bf8e82c84aed7/pathology/hovernet/prepare_patches.py#L20

My WSI is huge and could not be accommodated in .mat format. Also, I find this format to be unintutive to a python user. So, can we replace it with .npy source or wrap it to give a flexiblity to choose the best format?

[JHancox]

Hi @KumoLiu Thanks for your great work. I have been looking for the best hovernet implementation for my case. However, I have a comment here on using .mat format.https://github.com/KumoLiu/tutorials/blob/1b9f64d3a32fa08d37048dab282bf8e82c84aed7/pathology/hovernet/prepare_patches.py#L20

My WSI is huge and could not be accommodated in .mat format. Also, I find this format to be unintutive to a python user. So, can we replace it with .npy source or wrap it to give a flexiblity to choose the best format?

Hi @OmarAshkar - Thanks for your interest in Monai and Hovernet. So the .mat format is being used here to load annotations that are part of the ConSep dataset. Each .mat file will contain annotations for the nuclei. There should be no need to present your WSI in .mat format. If you want to run inference on your WSI, you don't need to use this script at all. If you do wish to train Hovernet on your own data, then you would need to prepare the data and annotations in a format of your choosing. You could certainly use .npy for that. Hope that helps.

[OmarAshkar]

Thanks @JHancox! yes I want to train my own data. I really wanted to extract the batches first instead of figuring out how to customize it for hovernet architecture. But I understand I think I will manipulate the script to my taste.

Thanks!