Here you will find a folder for each of the experiments that we worked on.
| Section | Description |
|---|---|
| tcxtc | Binary image classification of Tropical Cyclones and Extra-Tropical Cyclones. |
| multiclass | Image classification of Tropical Cyclones (i.e. Typhoons) in categories 2-5, according to labeling in JMA website. |
| pressure_regression | Regression from Tropical Cyclones images -> centre pressure values. |
In each of the folders, you will find the following content:
- notebooks: Folder containing some code snippets of the implemented experiment.
- predict.py: Script used to perform predictions on new data.
- weights.hdf5: Weights of the pre-trained model.
- preprocessing_x.h5: Parameters to preprocess the data (more details in folder).
We provide the weights of the model (weights.hdf5) and a script to do new
predictions (predict.py). Simply run
$ python predict.py weights.hdf5 <image_datafile.npy>
where image_datafile.npy is an image (or images) stored as a numpy array.
Accepted shapes are (256, 256) for a single image and (N, 256, 256) for a
batch of N images. In addition, you can use option -p to display
probabilities instead of labels. Find more details using --help.
Input image format for each of the experiments (i.e. models) may vary. Please review section Image format in the README file of the experiment you are interested in for more details.