Car Modernity Classification with Transfer Learning

This project applies transfer learning using a pre-trained ResNet-18 model to classify car modernity based on images. The goal is to categorize car designs by their production year and calculate a modernity score, which reflects the automotive design evolution.

Project Overview

The project focuses on using transfer learning to extract visual attributes from car images and predict their design modernity. The modernity score is a weighted sum of predicted probabilities for different year categories. The deep learning pipeline leverages a pre-trained ResNet-18 model on ImageNet and adapts it to classify cars into production year ranges.

Key Features:

• Transfer Learning: Fine-tuning a ResNet-18 model pre-trained on ImageNet.
• Modernity Score Calculation: Classifying car designs into production year categories and calculating a weighted modernity score.
• PyTorch Implementation: Using PyTorch for building and training the model.
• Dataset: DVM-CAR dataset, with car images labeled by production year.

Dataset

The dataset used in this project is the DVM-CAR dataset, which contains car images labeled with details such as brand, model, production year, and color. The images are categorized into five production year ranges:

• 0: 2000-2003
• 1: 2006-2008
• 2: 2009-2011
• 3: 2012-2014
• 4: 2015-2017

Data Preprocessing:

• Train/Validation/Test Split: The dataset is split into 70% training, 20% validation, and 10% test sets. Car models are kept intact within each split.
• Image Resizing: All images are resized to 224x224 pixels to match the input size expected by ResNet-18.
• Normalization: Images are normalized using the mean and standard deviation of ImageNet.
• Data Augmentation: Random cropping and resizing transformations are applied to the training dataset for better generalization.

Model Architecture

We use a ResNet-18 model pre-trained on ImageNet for feature extraction. Two transfer learning strategies are explored:

1. Fixed Feature Extractor: Using ResNet-18 as a fixed feature extractor and training only the final classification layer.
2. Fine-tuning: Fine-tuning the entire ResNet-18 model, along with the newly added classification layer.

Modernity Score Calculation:

The modernity score for a car is computed as the weighted sum of predicted probabilities across production year categories.