Muffin vs Chihuahua Detection #465

Muffin vs Chihuahua Detection/Dataset/README.md

@@ -0,0 +1,70 @@
+# Muffin vs Chihuahua Dataset
+
+## Overview
+
+This project uses the Muffin vs Chihuahua Image Classification Dataset to build deep learning models capable of distinguishing between muffins and chihuahuas from images.
+
+The dataset is designed to test the visual recognition capabilities of machine learning and deep learning models because muffins and chihuahuas often appear visually similar.
+
+## Dataset Source
+
+Kaggle Dataset:
+
+https://www.kaggle.com/datasets/samuelcortinhas/muffin-vs-chihuahua-image-classification
+
+## Dataset Structure
+
+Dataset contains two classes:
+
+* Chihuahua
+* Muffin
+
+The dataset is already divided into:
+
+* Train Set
+* Test Set
+
+## Dataset Statistics
+
+### Training Data
+
+* Chihuahua Images: 2559
+* Muffin Images: 2174
+
+### Testing Data
+
+* Chihuahua Images: 640
+* Muffin Images: 544
+
+### Total Images
+
+* Chihuahua: 3199
+* Muffin: 2718
+
+## Image Characteristics
+
+* RGB Images
+* Various image resolutions
+* Real-world image samples
+* Binary Image Classification Problem
+
+## Purpose
+
+The dataset is used to:
+
+* Perform Exploratory Data Analysis (EDA)
+* Apply Image Preprocessing
+* Train Deep Learning Models
+* Compare Multiple Architectures
+* Evaluate Classification Performance
+
+## Classes
+
+| Class Label | Description                      |
+| ----------- | -------------------------------- |
+| Chihuahua   | Images containing Chihuahua dogs |
+| Muffin      | Images containing Muffins        |
+
+## License
+
+Please refer to the original Kaggle dataset page for dataset licensing and usage terms.

Muffin vs Chihuahua Detection/Images/.gitkeep

@@ -0,0 +1 @@
+

Muffin vs Chihuahua Detection/Model/README.md

@@ -0,0 +1,210 @@
+# Muffin vs Chihuahua Detection using Deep Learning
+
+## Project Overview
+
+This project focuses on classifying images into two categories:
+
+* Chihuahua
+* Muffin
+
+Although visually different to humans in many cases, several muffin images resemble chihuahuas and vice versa, making this an interesting computer vision classification problem.
+
+The objective of this project is to compare multiple deep learning architectures and determine the best-performing model for the task.
+
+---
+
+# Objectives
+
+* Perform Exploratory Data Analysis (EDA)
+* Apply Image Preprocessing and Augmentation
+* Train Multiple Deep Learning Models
+* Evaluate Model Performance
+* Compare Results Across Architectures
+* Identify the Best Performing Model
+
+---
+
+# Exploratory Data Analysis (EDA)
+
+EDA was performed before model development.
+
+The following analyses were conducted:
+
+* Class Distribution Analysis
+* Sample Image Visualization
+* Dataset Statistics
+* Image Inspection
+
+Generated Visualizations:
+
+* sample_images.png
+* class_distribution.png
+
+---
+
+# Data Preprocessing
+
+The following preprocessing techniques were applied:
+
+* Image Resizing (224 × 224)
+* Pixel Value Normalization
+* Data Augmentation
+
+  * Rotation
+  * Width Shift
+  * Height Shift
+  * Zoom
+  * Horizontal Flip
+
+---
+
+# Models Implemented
+
+## 1. Custom CNN
+
+Architecture:
+
+* Conv2D
+* MaxPooling2D
+* Conv2D
+* MaxPooling2D
+* Conv2D
+* MaxPooling2D
+* Flatten
+* Dense Layer
+* Dropout
+* Output Layer
+
+### Results
+
+* Test Accuracy: 86.82%
+* Test Loss: 0.2967
+
+---
+
+## 2. MobileNetV2 (Transfer Learning)
+
+Transfer Learning was applied using the pretrained MobileNetV2 model.
+
+### Results
+
+* Test Accuracy: 99.32%
+* Test Loss: 0.0197
+
+---
+
+## 3. ResNet50 (Transfer Learning)
+
+Transfer Learning was applied using the pretrained ResNet50 model.
+
+### Results
+
+* Test Accuracy: 79.05%
+* Test Loss: 0.4630
+
+---
+
+## 4. EfficientNetB0 (Transfer Learning)
+
+Transfer Learning was applied using the pretrained EfficientNetB0 model.
+
+### Results
+
+* Test Accuracy: 45.95%
+* Test Loss: 0.7061
+
+---
+
+# Model Evaluation
+
+Evaluation metrics used:
+
+* Accuracy
+* Precision
+* Recall
+* F1-Score
+* Confusion Matrix
+
+Confusion Matrix Visualizations:
+
+* cnn_confusion_matrix.png
+* mobilenet_confusion_matrix.png
+* resnet50_confusion_matrix.png
+* efficientnet_confusion_matrix.png
+
+---
+
+# Model Comparison
+
+| Model          | Test Accuracy | Test Loss |
+| -------------- | ------------- | --------- |
+| Custom CNN     | 86.82%        | 0.2967    |
+| MobileNetV2    | 99.32%        | 0.0197    |
+| ResNet50       | 79.05%        | 0.4630    |
+| EfficientNetB0 | 45.95%        | 0.7061    |
+
+Visualization:
+
+* model_comparison.png
+
+---
+
+# Best Performing Model
+
+🏆 MobileNetV2 achieved the highest performance.
+
+### MobileNetV2 Results
+
+* Accuracy: 99.32%
+* Precision: 99%
+* Recall: 99%
+* F1 Score: 99%
+
+This demonstrates the effectiveness of transfer learning for image classification tasks.
+
+---
+
+# Folder Structure
+
+```text
+Muffin vs Chihuahua Detection
+│
+├── Dataset
+│   └── README.md
+│
+├── Images
+│   ├── sample_images.png
+│   ├── class_distribution.png
+│   ├── cnn_confusion_matrix.png
+│   ├── mobilenet_confusion_matrix.png
+│   ├── resnet50_confusion_matrix.png
+│   ├── efficientnet_confusion_matrix.png
+│   └── model_comparison.png
+│
+├── Model
+│   ├── muffin_vs_chihuahua.ipynb
+│   └── README.md
+│
+└── requirements.txt
+```
+
+---
+
+# How to Run
+
+1. Clone the repository.
+2. Install required dependencies.
+3. Download the dataset from Kaggle.
+4. Update dataset paths if necessary.
+5. Run the notebook sequentially.
+6. Train and evaluate the models.
+
+---
+
+# Conclusion
+
+Four deep learning models were trained and evaluated for Muffin vs Chihuahua classification.
+
+Among all tested architectures, MobileNetV2 produced the highest accuracy and the best overall classification performance. The results show that transfer learning significantly improves performance compared to a custom CNN for this dataset.
+
+This project demonstrates the practical application of deep learning and transfer learning techniques in image classification tasks.