An On-Going Framework for Easily Experimenting with Deep Learning Models for Bioimaging Analysis
Due to the broad use of deep learning methods in Bioimaging, it seems convenient to create a framework that facilitates the task of analysing different models and selecting the best one to solve each particular problem. In this work-in-progress, we are developing a Python framework to deal with such a task in the case of bioimage classification. Namely, the purpose of the framework is to automate and facilitate the process of choosing the best combination of feature extractors (obtained from transfer learning and other techniques), and classification models. The features and models to test are fixed by a simple configuration file to facilitate the use of the framework by non-expert users. The best model is automatically selected through a statistical study, and then it can be employed to predict the category of new images.
KeywordsDeep learning Machine learning Parallelization Bioimaging Image processing
- 3.Nauman, M., Ur Rehman, H., Politano, G., Benso, A.: Beyond Homology Transfer: Deep Learning for Automated Annotation of Proteins. bioRxiv (2017). https://www.biorxiv.org/content/early/2017/07/25/168120
- 6.Google: Google cloud automl (2018). https://cloud.google.com/automl/
- 8.Intel: Opencv (2016). https://opencv.org/
- 9.Chollet, F., et al.: Keras (2017). https://keras.io/