Usage
This page addresses how to train a new model and use an existing model to predict new data. Learn more about the required data structure here.
Learn more about required libraries to run this software here.
To train a new model, use train_model.py.
- Give a comma seperated path list for training data which needs to be completly annotated
- Give a comma seperated path list for validation data which ALSO needs to be completly annotated
- Define the CNN model as you wish
- Set saving paths for the trained model, its weights and plots
To predict data using a pre-trained model (see above), use predict_batch.py.
- Give the location of a saved model which should be used to predict on new data
- Give the directory which contains unannotated data that should be labeled
To make batch-prediction easy, you can use the built-in CLI. This function requires a pre-trained model and a batch of data, being located in a common root directory.
The CLI tool will prompt you to input all relevant variables and methods. Run the CLI interface using:
$ python predict_batch_custom.py
This is the recommended way of running predictions if you do not want to change the source-code.
