Realistic Image Classification

Project Description

This project involves classifying 80x80 images generated by deep generative models into three distinct classes (0, 1, 2).

Dataset

• Training Set: 10,500 images with labels
• Validation Set: 3,000 images with labels
• Test Set: 4,500 images

Approaches

KNN (K-Nearest Neighbors)

• Metrics Used: Minkowski, Euclidean, Manhattan
• Best Accuracy: 0.50 using Manhattan metric with 138 neighbors

CNN (Convolutional Neural Network)

• Framework: PyTorch
• Initial Accuracy: 70% on the validation set
• Final Model Accuracy: 76%

Image Processing

• Resize: 80x80x3 pixels
• Normalization: Pixel values scaled to [0, 1]

Data Augmentation

• Training Set: RandomHorizontalFlip, RandomRotation(5)
• Validation/Test Set: Only normalization

Model Architecture

1. Convolutional Layers: Conv2d, BatchNorm2d, ReLU, MaxPool2D
2. Dropout Layers: To prevent overfitting (rate: 0.1)
3. Dense Layers: Fully connected with ReLU activation, final layer with softmax activation

Hyperparameters

• Filters: {32, 64, 128, 256}
• Kernel Size: (3,3)
• Neurons in Dense Layers: {32, 64, 128, 256}
• Optimizer: Adam
• Loss Function: Cross entropy
• Early Stopping: Implemented to avoid overfitting

References

• Train CNN Model with PyTorch
• Image Classification Example
• PyTorch Documentation
• YouTube Tutorial
• FMI-Unibuc-IA