• 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

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

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

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

We provide the codes in ./ObjectDetection/, based on the mask-rcnn codebase

ResNet-50-XBNBlock-standard_train, ResNet-50-XBNBlock-advanced_train, [ResNeXt-50-XBNBlock-advanced_train