Welcome to the dedicated repository for advancing land cover recognition through the application of state-of-the-art models on satellite imagery. This repository serves as a comprehensive resource for researchers and practitioners in the field, providing access to research code, detailed setup instructions, and guidelines for conducting experiments with satellite image timeseries data.

This repository highlights contributions to the field through the following research publications:

• ViTs for SITS: Vision Transformers for Satellite Image Time Series - Featured at CVPR 2023, this paper explores the application of Vision Transformers to Satellite Image Time Series analysis. For further details, please consult the README_TSVIT.md document.
• Context-self contrastive pretraining for crop type semantic segmentation - Published in IEEE Transactions on Geoscience and Remote Sensing, this work introduces a novel supervised pretraining method for semantic segmentation of crop types exhibiti performance gains along object boundaries. Additional information is available in the README_CSCL.md document.

For the initial setup, please follow the instructions for downloading and installing Miniconda available at the official Conda documentation.

1. Creating the Environment: Navigate to the code directory in your terminal and create the environment using the provided .yml file by executing:

    conda env create -f deepsatmodels_env.yml
   

2. Activating the Environment: Activate the newly created environment with:

    source activate deepsatmodels
   

3. PyTorch Installation: Install the required version of PyTorch along with torchvision and torchaudio by running:

    conda install pytorch torchvision torchaudio cudatoolkit=10.1 -c pytorch-nightly
   

• Configuration: Specify the base directory and paths for training and evaluation datasets within the data/datasets.yaml file.
• Experiment Configuration: Use a distinct .yaml file for each experiment, located in the configs folder. These configuration files encapsulate default parameters aligned with those used in the featured research. Modify these .yaml files as necessary to accommodate custom datasets.
• Guidance on Experiments: For detailed instructions on setting up and conducting experiments, refer to the specific README.MD files associated with each paper or dataset.

This project is made available under the Apache License 2.0. Please see the LICENSE file for detailed licensing information.

Copyright © 2023 by Michail Tarasiou