resnet_attention

Implementation of Residual Attention Network for Image Classification Paper

Residual Attention Network Architecture

Implementation Details

Python Libraries

tensorflow               2.6.2
numpy                    1.19.5
matplotlib               3.5.0
python-magic             0.4.24

Other Dependencies

• Kaggle API for acquiring the dataset (pip install kaggle)
• For reference of the API click here

Hardware Used

GPU : NVIDIA RTX 3060

Dataset Used

• Weather Image Recognition (Kaggle Dataset by Jehan Bhathena)
• Image Classification Task with 11 Classes

References

• Residual Attention Network for Image Classification (Original Paper by Fei Wang et.al.)
• MixUp augmentation for image classification (Keras Code Examples article by Sayak Paul)

Training Parameters:

• Number of Training Epoch : 29
• Number of Batches: 32

Implementation Result

• Best Test Loss : 0.9973
• Best Test Accuracy : 67.71% (11 Classes)

Results of Training

Train/Validation Loss for 11 Classes of Weather Image

Train/Validation Accuracy for 11 Classes Weather Image