Image classification using machine learning methods

Recent advancements in deep learning have changed computer vision, evolving from traditional methods like Support Vector Machines (SVM) to Convolutional Neural Networks (CNN) in 2012, and more recently to Vision Transformers (ViT) in 2021. Simultaneously, improvements in computational power and the exponential growth of internet data have further driven this progress. This work uses these developments to build a computer vision model for classifying damaged and undamaged cars. Specifically, the study compares pre-trained convolutional neural networks, such as ResNet and ConvNeXt, to find the best-performing architecture for this task. Key contributions of this work include an analysis of the impact of data quantity and quality on model performance, the balance between computational resources and accuracy, and an exploration of hyperparameter optimization strategies for this application. The model's performance was evaluated using metrics like accuracy, model size, and F1 score, with ConvNeXt achieving a maximum accuracy of 95.4\%, demonstrating its effectiveness in binary classification for this task. The final model from this study could be applied to online platforms or auctions for selling used cars, potentially enhancing ad conversion rates, improving user experience, and reducing fraud associated with selling damaged vehicles.