Skip to content

Pytorch code for the paper "The color out of space: learning self-supervised representations for Earth Observation imagery"

License

Notifications You must be signed in to change notification settings

stevinc/TheColorOutOfSpace

Repository files navigation

The color out of space: learning self-supervised representations for Earth Observation imagery

This repository contains the PyTorch code for the paper:

The color out of space: learning self-supervised representations for Earth Observation imagery

Model architecture

Colorization & Multi-label classification - overview

Prerequisites

  • Python >= 3.7
  • PyTorch >= 1.5
  • CUDA 10.0

Dataset

We adopt the BigEarthNet Dataset. Refer to the README in the Colorization\dataset and Multi_label_classification\dataset folders for further information.

Models

Colorization

Different Encoder-Decoder combinations are available

  • Encoder ResNet18 - Decoder ResNet18
  • Encoder ResNet50 - Decoder ResNet50
  • Encoder ResNet50 - Decoder ResNet18

Multi Label Classification

The same encoders were employed in the colorization phase and an Ensemble model, composed of two equal encoders trained respectively on RGB and all other bands.

Training

Before running the files main.py contained in both the Colorization and Multi_label_classification folders you can set the desired parameters in the file job_config.py, which modify the ones contained in config/configuration.json.

Cite

If you have any questions, please contact [email protected], or open an issue on this repo.

If you find this repository useful for your research, please cite the following paper:

  @inproceedings{vincenzi2020color,
  title={The color out of space: learning self-supervised representations for Earth Observation imagery},
  author={Vincenzi, Stefano and Porrello, Angelo and Buzzega, Pietro and Cipriano, Marco and Pietro, Fronte and Roberto, Cuccu and 
          Carla, Ippoliti and Annamaria, Conte and Calderara, Simone},
  booktitle={25th International Conference on Pattern Recognition},
  year={2020}
}