Awesome

This is an official repo for paper "Explicitly Increasing Input Information Density for Vision Transformers on Small Datasets".

• version 1: Increasing input information density for vision transformers on small datasets, accepted as extended abstract by CVPR workshop (WiCV) 2022.
• version 2: Explicitly Increasing Input Information Density for Vision Transformers on Small Datasets, accepted by NeurIPS workshop (VTTA) 2022.

• main branch: for classification with vision transformers.
• heatmap branch: to select channels based on heatmaps.

1. Environment

pip install -r requirement.txt

2. Train

• Revise the data folder DATA_DIRin files under scripts_sh folder.
• Train using scripts in scripts_sh folder, e.g.

sh scripts_sh/swin_dct/dct/train_baseline_dct_flowers.sh

3. Test

• Checkpoints can be downloaded from Google drive
• Put checkpoints to corresponding folders and testing scripts are same with training.

4. Acknowledgments

Our codes are highly based on VT-drloc.