Towards Fine-grained Recognition: Joint Learning for Object Detection and Fine-grained Classification

Abstract

Fine-grained classification is a challenging problem due to subtle differences between intra-class categories. In practice, fine-grained classification is often used in conjunction with object detection algorithms to locate and identify object categories. Despite recent achievements in both fine-grained classification and object detection, few works have demonstrated datasets or solutions to simultaneously handle both tasks. We make two contributions to this problem. Firstly, we construct a fine-grained classification and detection benchmark. Secondly, we show an end-to-end convolutional neural networks (CNNs) architecture to detect and classify fine-grained objects. Experimental results verify that our networks perform favorably against alternatives.