Instructions for replication
This directory contains all the source code needed to reproduce this work.
Data preparation
To directly run the train and validation script in this directory, data preparation needs to be done. The
huggingface link contains all the training and synthetic testing data used in this work, stored in data.zip. This link also contains the pretrained model for single-phase and di-phase identification.
File single-phase_checkpoint_0200.pth and file bi-phase_checkpoint_2000.pth from the link above is the pretrained model, place them under directory "pretrained".
File data.zip contains the data and the annotaion file. Place directory "train" and "val" from data.zip under directory "data", place the annotation files anno_train.csv and anno_val.csv under directory "annotation".
Model Trianing
Single-phase
Run python train_single-phase.py to train the single-phase identification model from scratch. To train the model on your data, addtional parameters need to be set: python train_single-phase.py --data_dir_train=[your training data] --data_dir_val=[your validation data] --anno_train=[your anno file for training data] --anno_val=[your anno file for validation data].
Bi-phase
Run python train_bi-phase.py to train the bi-phase identification model. The bi-phase identification model is trained based on single-phase model, you can change the default setting by set the parameter load_path=[your pretrained single-phase model].
Model validation
Run python train_single-phase.py and python val_bi-phase.py to run the validation code at default setting.
If you wish to validate the model on your data, plase format your data using data_format.py
