[CVPR 2025] FasterPET (FPET)

The official pytorch implementation of FasterPET (FPET) from our paper Faster Parameter-Efficient Tuning with Token Redundancy Reduction

Kwonyoung Kim¹ Jungin Park^1* Jin Kim¹ Hyeongjun Kwon¹ Kwanghoon Sohn^1,2*

¹Yonsei University ²Korea Institute of Science and Technology (KIST) ^*Corresponding authors

Installation

conda create -n fpet python=3.10.13
conda install -r requirements.txt

Data Preparation

To download the datasets, please refer to https://github.com/ZhangYuanhan-AI/NOAH/#data-preparation. Set the directory of the dataset as <YOUR PATH>/fpet/data/.

Pretrained Model

Download the pretrained ViT-B/16 to <YOUR PATH>/fpet/ViT-B_16.npz.

Train & Evaluation

Train and evaluation on VTAB-1K.

sh train.sh

Citation

@inproceedings{kim2025faster,
  title={Faster Parameter-Efficient Tuning with Token Redundancy Reduction},
  author={Kim, Kwonyoung and Park, Jungin and Kim, Jin and Kwon, Hyeongjun and Sohn, Kwanghoon},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2025}
}

Acknowledgements

Our implementation is based on NOAH, timm, Binary Adapter and ToMe. Thanks for their awesome works.