👗 Neckline Classifier

The primary goal of this project is to classify different types of necklines in fashion images using deep learning models. The project includes:

CNN_MODEL_for_Neckline_Classification.ipynb

Implements a CNN architecture for classifying necklines:
🔹 This CNN consists of 4 convolutional layers, each followed by max pooling.
🔹 Uses two fully connected layers with dropout for regularization.
🔹 The final dense layer outputs 3 class probabilities using softmax activation.

📊 CNN Model Results

Confusion Matrix of CNN Model:

1. Label 0: Round Neck
2. Label 1: Scoop Neck
3. Label 2: V-Neck Neck

Neckline_Classification_Models_with_Oversampling.ipynb

Applied SMOTE to address class imbalance and improve model performance.
Implemented VGG16, MobileNet, and ResNet50 for feature extraction and classification.

📈 Model Performances

• VGG16 Confusion Matrix:

---

• MobileNet Confusion Matrix:

---

• Resnet50 Confusion Matrix:

---

🛠️ Used Technologies

Frameworks & Libraries

• TensorFlow – Deep learning framework for model building and training
• Keras – High-level API for designing neural networks
• OpenCV – Image preprocessing and manipulation
• Matplotlib / Seaborn – For visualizing data and results
• Scikit-learn – Confusion matrix, metrics, and SMOTE oversampling
• NumPy / Pandas – Data manipulation and numerical operations

Models Used

• Custom CNN (built from scratch)
• VGG16 (pretrained)
• MobileNet (pretrained)
• ResNet50 (pretrained)

Techniques

• SMOTE – Synthetic Minority Oversampling Technique for class imbalance
• Transfer Learning – Using pretrained models as feature extractors
• Image Normalization & Resizing – For consistent model input
• Softmax Activation – For multi-class classification