A Refined Training Recipe for Fine-Grained Visual Classification

Fine-grained visual classification (FGVC) aims to recognize subordinate categories within a super-category, such as specific car models. This task is challenging due to small inter-class variation and large intra-class variation from factors like pose and lighting. Instead of employing complex architectures, this approach focuses on developing a refined training recipe for a standard ResNet-50 model. By compounding modern training techniques and best practices, the goal is to achieve competitive performance on the Stanford Cars benchmark without modifying the core model architecture, demonstrating the power of a well-tuned training pipeline.