ResNet_Attention(CBAM，SE)

Official instruction: CBAM，SE

Required Environment

Ubuntu20.04
GTX 1080Ti
Python3.7
PyTorch1.7.0
CUDA10.2
CuDNN7.0

Usage Method(trian with CIFAR10)

The model's backbone is ResNet. In our training, we use CIFAR10 as our dataset.

# To train with SE
 python train_CIFAR10.py --prefix 4 --device 1 --epoch 160 --att_type se
# To trian with CBAM  
 python train_CIFAR10.py --prefix 5 --device 1 --epoch 160 --att_type cbam

Validation Result

• ResNet50 (trained for 160 epochs) ACC@1=93.41% ACC@5=99.84%
• ResNet50+SE (trained for 160 epochs) ACC@1=94.01% ACC@5=99.89%
• ResNet50+CBAN(trained for 160 epochs) ACC@1=93.41% ACC@5=99.90%

Result Graph

Blue：ResNet50

Red：ResNet50+SE