Evaluating the Performance of ResNet Variant Models for Car Model Detection Using a Transfer Learning Approach

This paper presents a comprehensive investigation into car model detection utilizing state-of-the-art deep learning architectures, specifically focusing on popular ResNet variants: ResNet18, ResNet50, and ResNet101. A dataset consisting of 6730 car images across 33 distinct classes was meticulously curated. Transfer learning using ResNet101 and meticulous training procedures resulted in a highly accurate model achieving an accuracy of 83.40%, outperforming other models including ResNet18 and ResNet50. Comparative analysis of various methods underscores the superiority of the proposed ResNet101-based approach in accurately identifying car models. The study highlights the potential and relevance of advanced deep learning architectures, offering promising applications in real-world scenarios for precise car model recognition. Future research will continue to refine and extend these advancements to further enhance performance and broaden the scope of car model detection.