Awesome
What's in this repository
You have reached the repository containing code to reproduce the experiments in our paper:
Test Time Transform Prediction for Open Set Histopathological Image Recognition
Adrian Galdran, Katherine Jane Hewitt, Narmin Ghaffari Laleh, Jakob Kather, Gustavo Carneiro, Miguel A. González Ballester
MICCAI - Medical Image Computing and Computer Assisted Interventions 2022
Link: here
Video Presentation
You can find watch a 5-minute video presentation we prepared also for MICCAI workshop by clicking in the image below:
<a href="https://www.youtube.com/watch?v=Dt9uAvgPPak"> <p align="center"> <img href="InstantDL" src="other/T3PO.png" width="500" alt="Link to presentation" align="center"> </p> </a>Prepare the data
Please follow these instructions to get the data ready:
- Kather 2016 link to data source
wget https://zenodo.org/record/53169/files/Kather_texture_2016_image_tiles_5000.zip
# below, optionally specify different path to store data, "-d other_path/"
unzip Kather_texture_2016_image_tiles_5000.zip -d data/Kather_texture_2016_image_tiles_5000
# then you need to use here that path, adding "--path_data_in other_path/Kather_texture_2016_image_tiles_5000"
python data/kather2016.py --path_data_in data/Kather_texture_2016_image_tiles_5000
rm -r data/Kather_texture_2016_image_tiles_5000
rm Kather_texture_2016_image_tiles_5000.zip
- Kather 100k link to data source
wget https://zenodo.org/record/1214456/files/NCT-CRC-HE-100K.zip
# below, optionally specify different path to store data, "-d other_path/"
unzip NCT-CRC-HE-100K.zip -d data/
# then you need to use here that path, adding "--path_data_in other_path/NCT-CRC-HE-100K"
python data/kather100k.py --path_data_in data/NCT-CRC-HE-100K
rm -r data/NCT-CRC-HE-100K
rm NCT-CRC-HE-100K.zip
Train the models
Afterwards, have a look at osr_train.sh, where you can find the instruction to train a model as in our paper.
Note that testing also happen in the training script, at the end, and results are already logged in a txt file.
Acknowledgement
Acknowledgement: The code in this repository is very much based on this codebase. In fact, you may find some pieces of code here and there that do nothing, those are parts of the original code that I did not get to remove or clean up, sorry about that.