Advance Convolutions, Attention and Image Augmentation: Depth wise, Pixel Shuffle, Dilated, Transpose, Channel Attention, and Albumentations Library
The dataset used here is CIFAR-10. It has the classes: ‘airplane’, ‘automobile’, ‘bird’, ‘cat’, ‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’. The images in CIFAR-10 are of size 3x32x32, i.e. 3-channel color images of 32x32 pixels in size.
The problem statement for this tutorial is given below
Figure 1 : Problem Statement
The summary of the model is shown below
Figure 2 : Model Summary
Code Explanation
The model.py file contains the class Net which defines the CNN.
Albumentation
The transformation is applied using library albumentations.
Figure 3 : Image showing applied transformations
Result
The training log for epoch 17 to 40 is shown below
Figure 4.1 : Log1
Figure 4.2 : Log2
Figure 4.3 : Log3
The test and validation loss and accuracy are shown below
Figure 5 : Loss and accuracy plots
Figure 6 : Validation accuracy per class
Misclassified Images during validation
Misclassified images during validation for all the three models are shown below
Figure 7 : Misclassified images