PSLT

Official Pytorch implementation of IEEE Transaction on Pattern Analysis and Machine Intelligence 2023, PSLT: A Light-weight Vision Transformer with Ladder Self-Attention and Progressive Shift. Please refer to paper and project page for more details. The optimized models can be found in the project page.

We currenently release the pytorch version code for:

• ImageNet-1k classification
• Object detection on COCO

Our repository is based on DeiT, timm, mmdet and PVT.

Requirements:

1. Image classification:
   • pytorch 1.9.0, torchvision 0.10.0 and timm 0.4.12
2. Object Detection
   • mmcv-full 1.4.0 and mmdet 2.22.0

Acknowledgement

This respository is built based on DeiT, timm, mmdet and PVT.

Citation

If you use this code for a paper, please cite:

@article{wu2023pslt,
title = {PSLT: A Light-weight Vision Transformer with Ladder Self-Attention and Progressive Shift},
author = {Gaojie Wu, Wei-Shi Zheng, Yutong Lu and Qi Tian},
journal = {IEEE Transaction on Pattern Analysis and Machine Intelligence},
year = {2023}
}