u-net_Segmentation/unet

• Unet based model training and inference
• Prerequisites
• Code structure
• • Training
  • Testing

Unet based model training and inference

This folder contains code for sustainbale farming practices detection using Unet.

Prerequisites

• Python 3.5+
• Imgaug
• Tensorflow 1.6+
• Keras 2.0.8+

Code structure

• main.py

This is the entry point, supports --runmode 'train' and 'test':

Training

python main.py --runmode=train --lossmode=bce --cudadev=0 --runprefix=tt3

Testing

python main.py --runmode=test --lossmode=bce --cudadev=0 --runprefix=tt3

• unet_config.py

Confiuration for training and testing the model. Here is the place to provide file path to the data, images dimentions, number of epocs and etc.

• trainer.py

Instantiates the model and trains for the specified number of steps\epochs.

• model.py

Here is definition of the Unet model.

• model_metrics.py

This file contains helper functions for model evaluation.

• losses.py

The place were custom loss functions (Dice coefficient based) reside.

• dataset_helper.py and dataset_reader.py

Here all data input related routines go.

• scorer.py

This file has heper functions to get predictions from the trained model.