A SimCLR implementation for self-supervised learning from unlabeled data
We use the SSL4EO-S12 (just the RGB channel) unlabeled dataset to learn the pretext task and for the down-stream task, we use a subset of the Eurosat (RGB channel) data. We use the simCLR based pipeline for the self-supervised learning task.
The following are some sample images from the SSL4EO-S12 RGB dataset.
The following are some sample images from the Eurosat RGB dataset which have been used for learning the down-stream task of classification.
The Google Colab (.ipynb) implementation of the same can be found in simCLR.ipynb file in this repository.