CatDogPandaClassifier

Table of Contents

1. Introduction
2. Dataset
3. Model Architecture
4. Training Pipeline
5. Results and Performance Metrics
6. Model Inference

---

1. Introduction

• This project implements a multi-class image classifier using a Convolutional Neural Network (CNN). The goal was to classify images of pandas, cats, and dogs. The dataset consists of 3000 images (1000 per class) and was split into an 80:20 ratio for training and validation. The final trained model achieved an accuracy of 91.33%.

• The project had the following purpose:

1. Building and Training a Neural Network from Scratch
   The project involves designing a CNN architecture, training it on real-world data, and analyzing its results.
2. Reinforcing a Complete Training Pipeline
   The project follows a robust training pipeline, addressing common errors such as:
   • Input/output shape mismatches.
   • Loss function and optimizer compatibility.
   • Image transformation issues.
   • Misaligned plotting curves.

---

2. Dataset

The dataset, obtained from Kaggle, consists of:

• 3000 images (1000 per class: Panda, Cat, Dog).
• Images resized to 224x224 for consistency.
• Data split into 80% training and 20% validation.
• Some data samples are visualized below
  
  • For Cats:
    
    
    
  • For Dogs:
    
    
    
  • For Pandas:
    
    
    

Preprocessing and Augmentation

To enhance generalization, the following augmentations were applied:

• Random rotation.
• Horizontal flipping.
• Random affine transfromation.
• Image Normalization

---

3. Model Architecture

For a detailed breakdown of the architecture, refer to Training_from_scratch.ipynb.

Key Highlights:

• Input: (Batch Size, 3, 224, 224)
• Output: 3-class scores (Panda, Cat, Dog).
• Activation Functions: ReLU for hidden layers, Softmax for output.
• Regularization: Dropout and batch normalization.

---

4. Training Pipeline

The training process involved:

• Experimentation with:
  • Number of layers and filters.
  • Optimizers and learning rate schedulers.
  • Regularization techniques like data augmentation, dropout, and batch normalization.
  • Number of epochs.
• Validation accuracy monitored to ensure the model generalizes well.

Explore the training pipeline in the following notebooks:

1. Data_and_Pipeline_Check.ipynb
2. Training_from_scratch.ipynb

---

5. Results and Performance Metrics

Validation Accuracy

The final validation accuracy achieved was 91.33%.

Confusion Matrix

The confusion matrix of the classification model is given below.

---

6. Model Inference

Here are some predictions made by the model:

---