This repository serves as a hub for resources, code, and explanations related to COVID-19 detection leveraging active learning. Active learning, a powerful machine learning paradigm, plays a pivotal role in optimizing the labeling process, enhancing model performance, and making the most of limited labeled data.
Showcases an active learning framework designed to significantly reduce the annotation burden in training COVID-19 detection models. Explores the synergy between machine learning and human expertise, where the model intelligently selects the most informative samples for labeling. Utilizes state-of-the-art deep learning architectures and image processing techniques for robust detection of COVID-19 cases from medical images. Offers a curated dataset containing diverse medical images for COVID-19, facilitating model training, validation, and experimentation. Encourages collaborative research and experimentation by providing a well-documented codebase and resources.
Virus
Normal
in Tensorflow-keras directory run this code:
python train.py
in Pytorch directory run this code:
python train_torch.py
in Active Learning (Pytorch) directory run this code:
python train_queries.py
You can change the hyperparameters in the main running file.
You can also add more models in the models directory using our importing format
You can use any datasets and using datasets directory files you can split and preprocess the datasets.
We achieved 98% validation accuracy using InceptionResnetV2 model and could reach around 0.0002 total validation loss in Active learning and normal classifiers in Pytorch and Tensorflow.