Unsupervised Coreset Selection

An implementation of core-set selection for image segmentation tasks.

Install dependecies

pip install -r requirements.txt

Usage

Run the following script to start training SimCLR and export coreset dataloader to disk:

python coreset-select.py

Note

• Your custom dataloader should implement the appropriate augmentation pipeline discussed in the SimCLR paper.
• This script also assumes you have the appropriate CUDA dependencies installed for training.

References

Ju, J. et al. (2021) Extending contrastive learning to unsupervised Coreset selection, arXiv.org. Available at: https://arxiv.org/abs/2103.03574 (Accessed: 01 February 2024).