Three self-attention modules based on three papers are implemented by PyTorch, which are:
Squeeze-and-Excitation Networks (resnet_se.py),
CBAM: Convolutional Block Attention Module (resnet_cbam.py), and
BAM: Bottleneck Attention Module (resnet_bam.py),
with the baseline being ResNet, proposed in
Deep Residual Learning for Image Recognition (resnet_base.py).