Official implementation of Representation Learning via Consistent Assignment of Views over Random Partitions (CARP)
Thirty-seventh Conference on Neural Information Processing Systems
Make sure to download CARP's pretraining files and place them in /pretrained/carp/ folder.
Checkpoints can be downloaded here.
| Method | Epochs | Multicrop | Top-1 | k-NN | URL |
|---|---|---|---|---|---|
| CARP | 100 | 2x224 + 6x96 | 72.5 | 63.5 | CARP-100ep |
| CARP | 200 | 2x224 + 6x96 | 74.2 | 66.5 | CARP-100ep |
| CARP | 400 | 2x224 | 73.0 | 67.6 | CARP-400ep |
| CARP | 400 | 2x224 + 6x96 | 75.3 | 67.7 | CARP-400ep |
To run evaluations, ensure you have a proper Python environment with PyTorch 2.0 and other dependencies.
Go to specific evaluation folders (such as knn or kmeans) for examples of how to run each.
Please, cite this work as:
@inproceedings{
Silva2023,
title={Representation Learning via Consistent Assignment of Views over Random Partitions},
author={Silva, Thalles and Ram\'irez Rivera, Ad\'in},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems ({NeurIPS})},
year={2023},
url={https://openreview.net/forum?id=fem6BIJkdv}
}