Deep Learning networks for segmentation.

LICENSE MIT:

     For full licence please refer to Unet/LICENSE

Short descritpion:

     Custom Unet implementations for image segmentation.

Aim:

     Provide a coherent library enlisting multiple well known Deep Learning architecutres for image Segmentation

How to use?

1. To train the model do check update the Unet/config.py
2. Run Unet/myTrain.py
3. Locate the weights under the weights/ directory
4. Load the model Unet/myModels.py
5. Load the weights using the standard Keras recipe
6. Enjoy.

Progress:

1. A custom UNET based approach has been included atm.

TODOs:

1. Reamend the layout
2. Package the Unet to be pip installable ... 

Project layout:

• Unet/
  • Unet/ → models, training, dataloading, deployement script (slurn based proprietary script)
  • demos/ inferecing and visualizing demos
  • data/ → empty dir to place data for training -- please update Unet/config.py
  • logs/ → directory required by slurm script for logging
  • weights/ → to store training weights
  • deps.txt → framwork dependencies

Depends on :

     The project do heavily depends on:

	1. Tensorflow >= 1.8
	2. Keras >=2.2.0
	3. Opencv 3.4.0
	4. Pandas, the notorious data "tabling" package

     For extra info about the python environment setup please refer to deps.txt. This is the entire virtual environment that I work on to build deep learning applications.