09 - Data Augmentation and Grad Cam

Session 9 - Data Augmentation and Grad Cam

The model reaches a maximum accuracy of 91.53% in 41 epochs on CIFAR-10 dataset using ResNet-18 model.

Gradient-weighted Class Activation Map (GradCAM) was implemented for each convolution block to generate model prediction heatmaps (Examples shown below).

Parameters and Hyperparameters

• Loss Function: Cross Entropy Loss (combination of nn.LogSoftmax and nn.NLLLoss)
• Optimizer: SGD
• Learning Rate: 0.01
• LR Step Size: 25
• LR Gamma: 0.1
• Batch Size: 64
• Epochs: 50

Data Augmentation

The following data augmentation techniques were applied to the dataset during training:

• Horizontal Flip
• Rotation
• CutOut

The albumentations package was used to apply augmentation.

GradCAM

Some of the examples where the network was focusing while predicting the output is shown below:

Image 1

Image 2

Image 3

Image 4

Change in Validation Loss and Accuracy

Correctly Classified Images

Misclassified Images

Project Setup

On Local System

Install the required packages
$ pip install -r requirements.txt

On Google Colab

Upload the files in the root folder and select Python 3 as the runtime type and GPU as the harware accelerator.

Group Members

• Shantanu Acharya (Canvas ID: 25180630)
• Rakhee (Canvas ID: 25180625)