This is the code repository for the publication "MyriadAL: Active Few Shot Learning for Histopathology", accepted to IEEE CAI 2024. The paper, details of the methodology, and results are available at https://arxiv.org/abs/2310.16161

First, install the requirements listed in the Python files. Then, set up directories nct_pickle/, lc25000_pickle/, and breakhis_pickle/, depending on the dataset of interest. Download the desired dataset from its respective online source, and pickle the dataset using the utilities in Pickle_Dataset/.

Then, pretrain the model using MoCo with the utilities in Pretrain_FSL_Model/, and generate the starting pseudo labels with Generate_Pseudo_Labels/.

Replace the paths in Active_FSL/base_main_moco_model.py and similar files, depending on the dataset of interest. Also modify the file config.py to suit your needs. Use the pretrained model and the generated pseudo labels in the appropriate spots.

The training process can then be started by running Active_FSL/base_main_moco_model.py. The results are stored in the log files located in the same folder, or outputted in the command line.