This project is the PyTorch implementation of our CVPR 2022 paper:

Requirements and Dependency

• Install PyTorch with CUDA (for GPU). (Experiments are validated on python 3.8.11 and pytorch 1.7.0)
• (For visualization if needed), install the dependency visdom by:

pip install visdom

===========================================================================

Note: I noted the pre-trained models are requested, here is the link of ResNet-XBNBlock-standard_train and ResNet-XBNBlock-advanced_train. I will upload our pre-trained ResNeXt model, if I can access the machine in my lab. (work at home due to COVID-19)

===========================================================================

Experiments

Here, we provide the code for reproducing the main experiments on ImageNet datasets.

1. Prepare the dataset:

Download the ImageNet-1K datasets, and put it in the dir: ./data/imageNet/ or you can specify your datapath by changing --dataset-root=/your-data-path

2. Run scripts of experiments:

We provide the scripts in ./experiments/, including the experiments on the ResNet, ResNeXt, Mobilenet-V2 and ShuffleNet-V2 .