Announcement

The code is modified from NOAH. We added necessary files and random seed to make the split generation reproducible.

Preparation

1. Get conda installed and run env_setup.sh
2. Download the required files and folders from this link. The link contains:
   • data: Move the this folder to the root directory of this repository. It contains necessary files to execute get_vtab1k.py
   • vtab-1k.tar.gz: VTAB-1k dataset processed by this repo, using seed 9526

Execute

To execute the script, run the following command:

python get_vtab1k.py --seed 9526

Publication

If you find this repo useful, please consider citing our work:

@misc{2409.16434,
    Author = {Zheda Mai and Ping Zhang and Cheng-Hao Tu and Hong-You Chen and Li Zhang and Wei-Lun Chao},
    Title = {Lessons Learned from a Unifying Empirical Study of Parameter-Efficient Transfer Learning (PETL) in Visual Recognition},
    Year = {2024},
    Eprint = {arXiv:2409.16434}
}