Multi-scale Deep Residual Networks for Fine-Grained Image Classification

Abstract

Fine-grained image classification aims at distinguishing very similar images, i.e., the subcategories in one class. Compared with generic object recognition, fine-grained image classification is much more challenging due to the small inter-class variance. Deep Residual Networks (ResNet) is a recently proposed deep Convolution Neural Networks (CNN) model, and has achieved the excellent performance on image classification. Though powerful, like other contemporary CNN models, ResNet only exploits the features extracted from the last output layer for classification, which may be insufficient for fine-grained classification. In this paper, we propose a Multi-scale Residual Networks (Multi-scale ResNet) to further improve the fine-grained image classification performance. Based on the ResNet model, we extract features from multiple CNN layers, add these high-level and mid-level features together with different weights for final classification. We compare our proposed model with some state-of-the-art models on two fine-grained image dataset, Stanford Cars and Dogs, and experimental results validate the efficacy of our method.