Home

Awesome

Pytorch implementation of TokenMix (ECCV 2022)

tenser

This repo is the offcial implementation of the paper TokenMix: Rethinking Image Mixing for Data Augmentation in Vision Transformers

@article{UniNet,
  author  = {Jihao Liu, Boxiao Liu, Hang Zhou, Yu Liu, Hongsheng Li},
  journal = {arXiv:2207.08409},
  title   = {TokenMix: Rethinking Image Mixing for Data Augmentation in Vision Transformers},
  year    = {2022},
}

Update

8/9/2022 Update the source code.

Preparation

Data

Following TokenLabeling to prepare ImageNet data and label maps generated with NFNet-F6.

Environment

The code is tested with torch==1.11 and timm==0.5.4.

Run experiments

Currently, we supporting running experiments with slurm. You can reproduce the results of Deit-small as follows:

sh exp/deit_small/run.sh partition

Models

ModelepochsTop-1 Acc.Ckpt
Deit-tiny30073.2-
Deit-small30080.8-
Deit-base30082.9-