This is a Keras implementation of "CBAM: Convolutional Block Attention Module". This repository includes the implementation of "Squeeze-and-Excitation Networks" as well, so that you can train and compare among base CNN model, base model with CBAM block and base model with SE block.

CBAM proposes an architectural unit called "Convolutional Block Attention Module" (CBAM) block to improve representation power by using attention mechanism: focusing on important features and supressing unnecessary ones. This research can be considered as a descendant and an improvement of "Squeeze-and-Excitation Networks".

• Python 3.x
• Keras

This repository use Cifar10 dataset. When you run the training script, the dataset will be automatically downloaded. (Note that you can not run Inception series model with Cifar10 dataset, since the smallest input size available in Inception series model is 139 when Cifar10 is 32. So, try to use Inception series model with other dataset.)

You can train and test base CNN model, base model with CBAM block and base model with SE block. You can run CBAM_block or SE_block added models in the below list.

• Inception V3 + CBAM / + SE
• Inception-ResNet-v2 + CBAM / + SE
• ResNet_v1 + CBAM / + SE (ResNet20, ResNet32, ResNet44, ResNet56, ResNet110, ResNet164, ResNet1001)
• ResNet_v2 + CBAM / + SE (ResNet20, ResNet56, ResNet110, ResNet164, ResNet1001)
• ResNeXt + CBAM / + SE
• MobileNet + CBAM / + SE
• DenseNet + CBAM / + SE (DenseNet121, DenseNet161, DenseNet169, DenseNet201, DenseNet264)

To change reduction ratio, you can set ratio on se_block and cbam_block method in models/attention_module.py

You can simply train a model with main.py.

1. Set a model you want to train.
   • e.g. model = resnet_v1.resnet_v1(input_shape=input_shape, depth=depth, attention_module=attention_module)
2. Set attention_module parameter
   • e.g. attention_module = 'cbam_block'
3. Set other parameter such as batch_size, epochs, data_augmentation and so on.
4. Run the main.py file
   • e.g. python main.py

• Blog: CBAM: Convolutional Block Attention Module
• Repository: CBAM-TensorFlow
• Repository: CBAM-TensorFlow-Slim
• Repository: SENet-TensorFlow-Slim

• Paper: CBAM: Convolutional Block Attention Module
• Paper: Squeeze-and-Excitation Networks
• Repository: Keras: Cifar10 ResNet example
• Repository: keras-squeeze-excite-network

Byung Soo Ko / [email protected]