Model Card for OpenSSL-SimCore

This repo contains some of the pretrained models from our paper, Coreset Sampling from Open-Set for Fine-Grained Self-Supervised Learning. We share the SimCore-pretrained models in the Open-set Self-Supervised Learning (OpenSSL) task, according to each fine-grained dataset. SimCore significantly improves representation learning performance in various downstream tasks, by leveraging a coreset sampled from the unlabeled open-set.

Model Details

Model Sources

• Repository: https://github.com/sungnyun/openssl-simcore
• Paper: Coreset Sampling from Open-Set for Fine-Grained Self-Supervised Learning, S. Kim et al., CVPR 2023

Model Type

SimCore with a stopping criterion (Table 2 in the main paper).

• Backbone: ResNet50
• SSL algorithm: SimCLR
• Open-set: ImageNet-1k
• Target dataset: aircraft, cars, pets, cub, dogs, flowers, stanford40, mit67, dtd, celeba, and food11

For use: Please check our github repo for the instructions.

License: Apache 2.0 License

Where to send questions or comments about the model: https://github.com/sungnyun/openssl-simcore/issues