Plant Network Segmentation

Welcome to the home of the Plant Network Segmentation code base. This code base can be used to interact with raw leaf images intended to be used as input to the Optical Vulnerability (OV) method.

The code base is broken into two components: a plant handler component and a TF2 model builder component.

The plant handler contains actions to interact with OV method data. In addition, it contains an object orientated data model to make further analysis easier. To interact with the plant handler use:

python -m src.__main__ -i

The TF2 model builder can be used to train a Tensorflow model. Three models are available for training, namely U-Net, U-Net (ResNet34), and W-Net. The trained model can be saved and then used to make predictions in the first component. However, the first component has been structured such that the prediction is independent of Tensorflow specifically. Any model which inherits the abstract Model class, and consequently implements a predict_tile method can be used. The how to use section explains how to interact with each component. To interact with the model builder use:

python -m src.pipelines.tensorflow_v2.__main__ -i

Models

The models available are shown below.

U-Net

U-Net (ResNet34)

W-Net

For more detailed documentation please visit: https://plant-network-segmentation.readthedocs.io/en/latest/