Awesome

Contrastive-Finetuning

This repo is the official implementation of the following paper: "On the Importance of Distractors for Few-Shot Classification" Paper

If you find this repo useful for your research, please consider citing this paper

@misc{das2021importance,
    title={On the Importance of Distractors for Few-Shot Classification},
    author={Rajshekhar Das and Yu-Xiong Wang and JoséM. F. Moura},
    year={2021},
    eprint={2109.09883},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

Dataset Download

To set up the dataset, follow the exact steps outlined in here.

Pretrained Model

To download the pretrained backbone model, follow the exact steps outlined in here

Running

• To run contrastive finetuning on cub data (default target domain) with the downloaded pretrained model, simply run bash conft.sh
• To run the multi-task variant on the same target domain, run bash mt_conft.sh
• To change the target domain or other hyperparameters, refer to conft.sh and mt_conft.sh

Acknowlegements

Part of the codebase, namely, the dataloaders have been adapted from Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation.