Master's Thesis in Ukrainian Catholic University (2018)
All the details on the data, preprocessing, model architecture and training details can be found in thesis text.
Experiment weights can be downloaded from Google Drive.
Requirements are specified in requirements.txt.
pip install -r requirements.txtModel works only on cuda.
There are two main scripts - trainer.py and evaluate.py, which are used for training and evaluation.
python trainer.py {experiment_name}python evaluate.py {experiment_name} ({dataset_name})All parameters are specified in config.yaml file in corresponding experiment folder.
Examples:
python trainer.py e010python evaluate.py e010 dexter+objectFolders dataset, model, criterion, metric, optimizer contain datasets, models, losses, metrics for evaluation and optimizers, respectively.
To get a specific dataset, model, loss, metric or optimizer, call functions get_dataloder, get_model, get_criterion, get_metric or get_optimizer, respectively. Functions are defined in __init__.py files in corresponding folders.