Contents

• Contents
  • Aug-ViT Description
  • Model architecture
  • Dataset
  • Environment Requirements
  • Script description
    • Script and sample code
  • Eval process
    • Usage
    • Launch
    • Result
  • Description of Random Situation
  • ModelZoo Homepage

Aug-ViT Description

Aug-ViT inserts additional paths with learnable parameters in parallel on the original shortcuts for alleviating the feature collapse. The block-circulant projection is used to implement augmented shortcut, which brings negligible increase of computational cost.

Paper: Yehui Tang, Kai Han, Chang Xu, An Xiao, Yiping Deng, Chao Xu, Yunhe Wang. Augmented Shortcuts for Vision Transformers. NeurIPS 2021.

Model architecture

A block of Aug-ViT is show below:

Dataset

Dataset used: CIFAR-10

• Dataset size: 60000 colorful images in 10 classes
  • Train: 50000 images
  • Test: 10000 images
• Data format: RGB images.

Environment Requirements

• Hardware(Ascend/GPU)
  • Prepare hardware environment with Ascend or GPU.
• Framework
  • MindSpore
• For more information, please check the resources below£º
  • MindSpore Tutorials
  • MindSpore Python API

Script description

Script and sample code

AugViT
├── eval.py # inference entry
├── fig
│   └── augvit.png # the illustration of augvit network
├── readme.md # Readme
└── src
    ├── config.py # config of model and data
    ├── c10_dataset.py # dataset loader
    └── augvit.py # augvit network

Eval process

Usage

After installing MindSpore via the official website, you can start evaluation as follows:

Launch

# infer example
  GPU: python eval.py --model augvit_s --dataset_path cifar_dataset --platform GPU --checkpoint_path [CHECKPOINT_PATH]

checkpoint can be downloaded at https://download.mindspore.cn/model_zoo/research/cv/augvit/.

Result

result: {'acc': 0.98} ckpt= ./augvit_c10.ckpt

Description of Random Situation

In dataset.py, we set the seed inside "create_dataset" function. We also use random seed in train.py.

ModelZoo Homepage

Please check the official homepage.