GeoSSL

This repository contains the code for the paper "Self-Supervised Learning for Scene Classification in Remote Sensing: current State of the Art and Perspectives".

Implemented Methods

• SimCLR[1]
• Momentum Contrast (v2)[2] (MoCo v2)
• BarlowTwins[3]
• Bootstrap Your Own Latent[4] (BYOL)

Installing

geossl was tested with and has the following dependencies:

python>=3.7
pytorch=1.10.2+cu102
torchvision=0.11.3+cu102
torchgeo=0.3.0

Exact dependencies are specified in the environment.yml file and it can be used to create an associated conda environment:

conda env create --file ./environment.yml

Usage

The train_evaluate.py script perform a self-supervised pre-training and evaluate the pre-trained model on the linear-evaluation protocol and fine-tuning (with 1% and 10% respectively). An example usage is available in scripts/train_evaluate.sh.

Pre-trained weights

The following pre-trained weights are available to download:

Backbone	Pre-training dataset	Method	Identifier
ResNet18	RESISC45	SimCLR	"resnet18/resisc45/simclr"
ResNet18	RESISC45	MoCo v2	"resnet18/resisc45/moco"
ResNet18	RESISC45	BYOL	"resnet18/resisc45/byol"
ResNet18	RESISC45	Barlow Twins	"resnet18/resisc45/barlow"
ResNet18	EuroSAT	SimCLR	"resnet18/eurosat/simclr"
ResNet18	EuroSAT	MoCo v2	"resnet18/eurosat/moco"
ResNet18	EuroSAT	BYOL	"resnet18/eurosat/byol"
ResNet18	EuroSAT	Barlow Twins	"resnet18/eurosat/barlow"

There is an helper on the backbone class to instantiate the model with the pre-trained weights:

from geossl.backbones import ResNetBackbone

model = ResNetBackbone.from_pretrained("resnet18/eurosat/simclr")

Citing

If this work is useful to your research, consider citing the paper associated with this code: "Self-Supervised Learning for Scene Classification in Remote Sensing: current State of the Art and Perspectives".

@Article{rs14163995,
	author = {Berg, Paul and Pham, Minh-Tan and Courty, Nicolas},
	title = {Self-Supervised Learning for Scene Classification in Remote Sensing: Current State of the Art and Perspectives},
	journal = {Remote Sensing},
	volume = {14},
	year = {2022},
	number = {16},
	article-number = {3995},
	url = {https://www.mdpi.com/2072-4292/14/16/3995},
	issn = {2072-4292},
	doi = {10.3390/rs14163995}
}

References

[1]Chen, T., Kornblith, S., Norouzi, M., & Hinton, G. (2020, November). A simple framework for contrastive learning of visual representations. In International conference on machine learning (pp. 1597-1607). PMLR.

[2]Chen, X., Fan, H., Girshick, R., & He, K. (2020). Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297.

[3]Zbontar, J., Jing, L., Misra, I., LeCun, Y., & Deny, S. (2021, July). Barlow twins: Self-supervised learning via redundancy reduction. In International Conference on Machine Learning (pp. 12310-12320). PMLR.

[4]Grill, J. B., Strub, F., Altché, F., Tallec, C., Richemond, P., Buchatskaya, E., ... & Valko, M. (2020). Bootstrap your own latent-a new approach to self-supervised learning. Advances in Neural Information Processing Systems, 33, 21271-21284.