---
license: cc-by-nc-4.0
datasets:
- abo
- inaturalist
pipeline_tag: image-feature-extraction
---


## Model training and evaluation

Instructions for how to train and evaluate a MILE model, as well as the necessary code are in the [Amazon Science repository](https://github.com/amazon-science/mile).

## Intended use and limitations

These models have been trained on [Amazon-Berkeley-Objects](https://amazon-berkeley-objects.s3.amazonaws.com/index.html) and [iNaturalist](https://github.com/visipedia/inat_comp/tree/master/2017) and are intended to demonstrate the power of the object-level embeddings w.r.t. the object /category retrieval task. For other domains or tasks, it should be further fine-tuned on relevant data.

 * [ABO](https://huggingface.co/AmazonScience/MILE/resolve/main/blrp-dinov2-vitl14-reg-lora0-fw-5e-06-mt09996-bs3-800_ckpt0390.pth)
 * [iNaturalist](https://huggingface.co/AmazonScience/MILE/resolve/main/mile-vitl14-reg-LRx8-g48x8-100M-mt0996-bs3-s250000-ep800-1_ckpt0361.pth)


## Citation
If you use this work, please cite:

```
@inproceedings{leotescu2024mile,
  title={Self-Supervised Incremental Learning of Object Representations from Arbitrary Image Sets},
  author={Leotescu, George and Popa, Alin-Ionut Popa and Grigore, Diana and Voinea, Daniel and Perona, Pietro},
  booktitle={Proceedings of WACV},
  year={2025}
}
```

## License

This library is licensed under the CC BY NC License.