This repo is the official implementation of the Gist framework.

• Create a conda virtual environment and activate it:

conda create -n gist python=3.8 -y
conda activate gist

• Install requirements:

pip install -r requirements.txt

• VTAB-1K

You can follow ssf ("Scaling & Shifting Your Features: A New Baseline for Efficient Model Tuning") to download them.

• For pre-trained ViT-B/16 on ImageNet-21K, the model weights will be automatically downloaded. You can also manually download them from ViT.

To fine-tune a pre-trained ViT model via Adapter within our GIST framework on VTAB-1K, run:

bash train_scripts/vit/train_vtab.sh

The code is built upon ssf.