A tutorial on AI to solve the problem of object detection and segmentation using Keras. We use a modified U-Net to perform a basic instance segmentation on 128px targets.

You can find the entire process of data preperation, model definition, training, and evaluation all in one Google Colab file here.

To generate the dataset for training or testing you first need to run:

python prepare_dataset.py

Once the datasets are ready, train the model by running:

python train.py

You can generate novel random scences by using the same functions in the prepare_dataset.py. Then you can either a command line or GUI testing application by running:

python test_gui.py

or:

python test.py

The file environment.yml is a conda environment file. The dataset.py script loads and resample the data to the desired resolutoin.