Inside the partA and partB directory i have provided README files for each train.py and ipynb file kindly use that in respective directory partA or partB.

The below is for only ipynb files

This repository contains two Jupyter Notebook files for building Convolutional Neural Network (CNN) architectures for image classification tasks. The notebooks are divided into two parts: Part A and Part B.

Part A: CNN Architecture from Scratch

• Usage:

  • Pass the path of train dataset folder in Question 2 "arguments" method and in Question 4 "arguments_2" method :

  • Pass the path of test dataset folder in Question 4 "arguments_2" method :

The architecture can be customized by adjusting parameters by applying these parameters in sweep such as:

• Number of Filters: 32, 64, 128
• Activation Function: ReLU, GELU, SiLU, Mish
• Filter Organization: Same, Double, Halve
• Batch Normalization: Yes, No
• Dropout Value: 0.2, 0.3
• Learning Rate: 0.001, 0.0001
• Number of Epochs: 5, 10
• Number of Neurons in Dense Layer: 128, 256, 512, 1024
• Batch Size: 32, 64
• Data Augmentation: Yes, No

Part B: Pre Trained CNN Model and Fine-Tuning

• Usage:

  • Pass the path of train dataset folder in "arguments" method

  • As i am applying pre-trained model on train dataset by spliting it in 80% training and 20% validation i am fine tuning the pre-trained Resnet model for better validation accuracy.

In Part B, the following parameters can be customized:

• Number of filters: 32, 64, 128

• Activation function: ReLU, GELU, SiLU, Mish

• Filter organization: Same, Double, Halve

• Batch normalization: Yes, No

• Dropout value: 0.2, 0.3

• Learning rate: 0.001, 0.0001

• Number of epochs: 5, 10

• Number of neurons in dense layer: 128, 256, 512, 1024

• Batch size: 32, 64

• Freezing option: All Except Last, Except First, Up To

• Freeze index: 2, 3, 4

• Data augmentation: Yes, No

• Note:

  • Run Each cell Sequencially Top to Bottom for avoiding errors.

  • Each notebook provides detailed instructions and code comments to guide customization and execution.

  • It is recommended to have a GPU-enabled environment for faster training.

  • Ensure that the dataset folders are correctly structured and contain the necessary images for training and testing.

Feel free to explore and experiment with different parameter configurations to achieve optimal performance for your image classification tasks!