Merge pull request #1012 from KrishnaChaitanya16/main

Hair Type Classification using Deep Learning v1

Author: abhisheks008

Hair Type Classifier/Dataset/Readme.md

@@ -0,0 +1,23 @@
+This is  an image dataset designed to classify various hair types. It includes high-quality images of individuals with diverse hair types. The dataset is helpful for training machine learning models to recognize and classify hair types.
+
+The dataset includes the major 4 types of hair and dreadlocks.
+
+-   Straight
+-   Wavy
+-   Curly
+-   Kinky
+-   Dreadlocks (not actually a hair type but a hair style)
+## Data Distrubution
+| Hair Type  | Training Images | Testing Images |
+|------------|----------------|---------------|
+| Kinky      | 173 | 44 |
+| Dreadlocks | 354            | 89 |
+| Straight   | 390 | 98 |
+| Curly      | 411 | 103 |
+| Wavy       | 264 | 66 |
+
+---
+## Dataset Link
+Here is the dataset link : [Link to the dataset](__https://www.kaggle.com/datasets/kavyasreeb/hair-type-dataset__).
+
+Download the dataset from the above link place it in Dataset Folder and make sure each class has its folder.

Hair Type Classifier/Images/ConvNext-tiny.png

@@ -0,0 +1,115 @@
+### Hair Type Classification Model using Deep Learning ###
+
+
+## Goal
+The main objective of this project is to develop a deep learning-based hair type classification system that can accurately classify hair into five categories: Kinky, Dreadlocks, Straight, Curly, and Wavy. This project aims to enhance automated hair classification accuracy using an ensemble of three powerful deep learning models.
+
+## Overview
+Hair type classification is a crucial aspect in various domains, including cosmetics, fashion, and dermatology. The goal of this project is to build an **AI-powered hair type classification system** that can categorize hair into five types:  
+- **Kinky**
+- **Dreadlocks**
+- **Straight**
+- **Curly**
+- **Wavy**  
+## Objectives
+- Develop a **deep learning-based model** for hair type classification.  
+- Train and evaluate multiple models, then combine them using an **ensemble approach**.  
+- Achieve **higher accuracy** through a hybrid model combining **ResNet50, VGG16, and ConvNeXt-Tiny**.  
+
+## Dataset
+Here is the dataset link : [Link to the dataset](https://www.kaggle.com/datasets/kavyasreeb/hair-type-dataset)
+## Data Distrubution
+| Hair Type  | Training Images | Testing Images |
+|------------|----------------|---------------|
+| Kinky      | 173            | 44            |
+| Dreadlocks | 354            | 89            |
+| Straight   | 390            | 98            |
+| Curly      | 411            | 103           |
+| Wavy       | 264            | 66            |
+
+
+
+---
+## Methodology
+- **Image Resizing**: All images resized to **224x224** for model compatibility.  
+- **Normalization**: Applied using **ImageNet statistics**.  
+- **Data Augmentation**: Used **Albumentations** for better generalization.  
+
+## Model Selection & Training
+I implemented and trained following three models:
+1. **ResNet50** – A deep residual network for effective feature extraction.  
+2. **VGG16** – A deep CNN with uniform kernel sizes.  
+3. **ConvNeXt-Tiny** – A modern CNN inspired by Vision Transformers. 
+### Ensemble Learning Approach  
+The **final model** is an **ensemble of all three models**, using a **weighted averaging technique** to achieve **higher accuracy**.  
+## Results
+### Classification Report for **Hair-ResNet50**:
+
+| Class       | Precision | Recall | F1-Score | Support |
+|-------------|-----------|--------|----------|---------|
+| Straight    | 0.92      | 0.92   | 0.92     | 195     |
+| Wavy        | 0.82      | 0.81   | 0.82     | 126     |
+| Curly       | 0.86      | 0.92   | 0.89     | 186     |
+| Dreadlocks  | 0.97      | 0.99   | 0.98     | 193     |
+| Kinky       | 0.91      | 0.77   | 0.83     | 95      |
+
+- **Accuracy**: 0.90 (795 samples)
+- **Macro Average**: Precision 0.90, Recall 0.88, F1-Score 0.89
+- **Weighted Average**: Precision 0.90, Recall 0.90, F1-Score 0.90
+### Classification Report for Hair-VGG16:
+
+| Class       | Precision | Recall | F1-Score | Support |
+|-------------|-----------|--------|----------|---------|
+| Straight    | 0.88      | 0.94   | 0.91     | 195     |
+| Wavy        | 0.89      | 0.83   | 0.86     | 126     |
+| Curly       | 0.91      | 0.92   | 0.91     | 186     |
+| Dreadlocks  | 0.98      | 0.97   | 0.98     | 193     |
+| Kinky       | 0.90      | 0.85   | 0.88     | 95      |
+
+- **Accuracy**: 0.92 (795 samples)
+- **Macro Average**: Precision 0.91, Recall 0.90, F1-Score 0.91
+- **Weighted Average**: Precision 0.92, Recall 0.92, F1-Score 0.92
+### Classification Report for hair-convnext-tiny:
+
+| Class       | Precision | Recall | F1-Score | Support |
+|-------------|-----------|--------|----------|---------|
+| Straight    | 0.92      | 0.96   | 0.94     | 195     |
+| Wavy        | 0.87      | 0.82   | 0.84     | 126     |
+| Curly       | 0.89      | 0.95   | 0.92     | 186     |
+| Dreadlocks  | 0.98      | 0.98   | 0.98     | 193     |
+| Kinky       | 0.94      | 0.80   | 0.86     | 95      |
+
+- **Accuracy**: 0.92 (795 samples)
+- **Macro Average**: Precision 0.92, Recall 0.90, F1-Score 0.91
+- **Weighted Average**: Precision 0.92, Recall 0.92, F1-Score 0.92
+### Ensemble Model Classification Report:
+
+| Class       | Precision | Recall | F1-Score | Support |
+|-------------|-----------|--------|----------|---------|
+| Straight    | 0.92      | 0.94   | 0.93     | 195     |
+| Wavy        | 0.88      | 0.85   | 0.86     | 126     |
+| Curly       | 0.91      | 0.93   | 0.92     | 186     |
+| Dreadlocks  | 1.00      | 1.00   | 1.00     | 193     |
+| Kinky       | 0.90      | 0.86   | 0.88     | 95      |
+
+- **Accuracy**: 0.93 (795 samples)
+- **Macro Average**: Precision 0.92, Recall 0.92, F1-Score 0.92
+- **Weighted Average**: Precision 0.93, Recall 0.93, F1-Score 0.93
+## Key Observations  
+- **Ensemble Model** provided the best results (**93.7% accuracy**).  
+- **ConvNeXt-Tiny** performed better than ResNet50 and VGG16.  
+- **Data Augmentation** significantly improved model robustness. 
+## Required Libraries
+- fastai  
+- fastai.vision.all  
+- fastai.callback.tracker  
+- albumentations -- version 1.3.0 
+- sklearn.metrics  
+- cv2 (OpenCV)  
+- pathlib
+## Note 
+Make sure the input and output directory paths and libraries are correctly set before running the code.
+
+
+
+

Hair Type Classifier/Images/Convnext-tiny_Confusion.png

@@ -0,0 +1,7 @@
+fastai
+albumentations==1.3.0
+scikit-learn
+opencv-python
+pathlib
+numpy
+pandas