SemiNMF-Autoencoders

This repository contains the code for reproducing the experiments in Semi-NMF Regularization-Based Autoencoder Training for Hyperspectral Unmixing. Design of the semi-NMF objective can be found in Algorithm.pdf.

Dependencies

• Docker 19.03.12
• PyTorch 1.9.0
• Python 3.7.10

Installation

Data

The dataset is publicly available and can be found here.
Download the Samson dataset from the above-mentioned source. Follow the directory tree given below:

|-- [root] HyperspecAE\
    |-- [DIR] data\
        |-- [DIR] Samson\
             |-- [DIR] Data_Matlab\
                 |-- samson_1.mat
             |-- [DIR] GroundTruth
                 |-- end3.mat
                 |-- end3_Abundances.fig
                 |-- end3_Materials.fig

From Docker (Recommended)

Using a docker image requires an NVIDIA GPU. If you do not have a GPU please follow the directions for installing from source. In order to get GPU support you will have to use the nvidia-docker2 plugin. The docker image is cached on the GPU with id 0. In case of OOM errors at training, pass two GPUs.

# Build the Dockerfile to create a Docker image.
docker build -t dgoel04/snreg:1.0 .

# This will create a container from the image we just created.
docker run -it --gpus '"device=gpu-ids"' dgoel04/snreg:1.0

From source:

1. Install the data by following the steps shown under installation.

2. Clone this repository.
   git clone https://github.com/dv-fenix/SemiNMF-Autoencoders.git
cd SemiNMF-Autoencoders

3. Install the requirements given in requirements.txt.
   python -m pip install -r requirements.txt

4. Change working directory.
   cd run

Run Experiments

Training the Autoencoders

The code is fairly modular and can be run from the terminal.

# For more information on the optional experimental setups and configurations.
python ../src/train.py --help

# You can manually change the arguments in samson_train.sh to choose the different autoencoder configurations.
sh samson_train.sh

Please make sure that all the arguments are to your liking before getting started with the training!

Abundance Map and End-Member Extraction

Please ensure that the arguments contained within extract.sh match those used in samson_train.sh during training.

# You can manually change the arguments in experiments.sh to choose the different configurations.
sh extract.sh

Cite

If you use this code, please cite our paper:

@inproceedings{goel2024semi,
  title={Semi-NMF Regularization-Based Autoencoder Training for Hyperspectral Unmixing},
  author={Goel, Divyam and Khanna, Saurabh},
  booktitle={2024 National Conference on Communications (NCC)},
  pages={1--6},
  year={2024},
  organization={IEEE}
}