SENet.pytorch

An implementation of SENet, proposed in Squeeze-and-Excitation Networks by Jie Hu, Li Shen and Gang Sun, who are the winners of ILSVRC 2017 classification competition.

Now SE-ResNet (18, 34, 50, 101, 152/20, 32) and SE-Inception-v3 are implemented.

• python cifar.py runs SE-ResNet20 with Cifar10 dataset.

• python imagenet.py IMAGENET_ROOT runs SE-ResNet50 with ImageNet(2012) dataset.

  • You need to prepare dataset by yourself
  • First download files and then follow the instruction.
  • The number of GPUs and workers, the learning rate is fixed so check and change them if needed.

For SE-Inception-v3, the input size is required to be 299x299 as original Inception.

Result

SE-ResNet20/Cifar10

python cifar.py [--baseline]

SE-ResNet50/ImageNet

The initial learning rate and mini-batch size are different from the original version because of my computational resource (0.6 to 0.1 and 1024 to 128 respectively).

• (*): He, K., Zhang, X., Ren, S., & Sun, J. (2015). Deep Residual Learning for Image Recognition.

• (**): I share this weight (training after 100 epochs).

senet = se_resnet50(num_classes=1000)
senet.load_state_dict(torch.load("weight.pkl"))

References

paper

authors' Caffe implementation