Coarse2Fine: a two-stage training method for fine-grained visual classification

Small inter-class and large intra-class variations are the key challenges in fine-grained visual classification. Objects from different classes share visually similar structures, objects same class can have poses viewpoints. Therefore, proper extraction of discriminative local features (e.g., bird’s beak or car’s headlight) is crucial. Most recent successes on this problem based upon attention models which localize attend parts. In work, we propose a training method for networks, Coarse2Fine, creates differentiable path attended feature maps to input space. Coarse2Fine learns an inverse mapping function informative regions raw image, will guide better features. Besides, initialization weights. Our experiments show that reduces classification error by up 5.1% common datasets.