SENet-pytorch

Pytorch implementation of SENet("Squeeze-and-Excitation Networks", 2017) based on resnet.

To-do: add another backbone model.(Inception).

Usage

To train the model, run:

python main.py --arch resnet50 --datasets CIFAR100 --n_epochs 100

Common arguments:

• --arch: Backbone model (resnet18, resnet34, resnet50, resnet101, resnet152)
• --datasets: Dataset to use (CIFAR10, CIFAR100, ImageNet)
• --use_senet: Enable SENet module (True/False, default: True)
• --reduction_ratio: Reduction ratio for the SE block (default: 16)