Awesome

Deep Generative Reflectance Fusion

Achieving Landsat-like reflectance at any date by fusing Landsat and MODIS surface reflectance with deep generative models.

Getting Started

Running experiments

Setup YAML configuration files specifying experiment : dataset, model, optimizer, experiment. See here for examples.

Execute training on, say GPU 0, as:

$ python run_training.py --cfg=path/to/config.yaml --o=output/directory --device=0

Once training completed, specify model checkpoint to evaluate in previously defined YAML configuration file and run evaluation as:

$ python run_testing.py --cfg=path/to/config.yaml --o=output/directory --device=0

Preimplemented experiments

Compile ESTARFM

To compile ESTARFM please follow guidelines from official repository.

Project Structure

├── data/
├── repro/
├── src/
│   ├── cuESTARFM
│   ├── deep_reflectance_fusion
│   ├── prepare_data
│   └── utils
├── tests
├── run_training.py
├── run_testing.py
├── run_ESTARFM.py
└── run_ESTARFM_evaluation.py

Directories :

• data/ : Landsat-MODIS reflectance time series dataset and experiments outputs
• repro/: bash scripts to run data version control pipelines
• src/: modules to run reflectance patches extraction and deep reflectance fusion experiments
• tests/: unit testing
• utils/: miscellaneous utilities

Installation

Code implemented in Python 3.8

Setting up environment

Clone and go to repository

$ git clone https://github.com/Cervest/ds-generative-reflectance-fusion.git
$ cd ds-generative-reflectance-fusion

Create and activate environment

$ pyenv virtualenv 3.8.2 fusion
$ pyenv activate fusion
$ (fusion)

Install dependencies

$ (fusion) pip install -r requirements.txt

Setting up dvc

From the environment and root project directory, you first need to build symlinks to data directories as:

$ (fusion) dvc init -q
$ (fusion) python repro/dvc.py --link=where/data/stored --cache=where/cache/stored

if no --link specified, data will be stored by default into data/ directory and default cache is .dvc/cache.

To reproduce a pipeline stage, execute:

$ (fusion) dvc repro -s stage_name

In case pipeline is broken, hidden bash files are provided under repro directory

References

@misc{bouabid2020predicting,
      title={Predicting Landsat Reflectance with Deep Generative Fusion}, 
      author={Shahine Bouabid and Maxim Chernetskiy and Maxime Rischard and Jevgenij Gamper},
      year={2020},
      eprint={2011.04762},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}