[8] IU-Module: Intersection and Union Module for Fine-Grained Visual Classification

Published in July 06, 2020

Recommended citation: Yixiao Zheng, Dongliang Chang, Jiyang Xie, and Zhanyu Ma\*. IU-Module: Intersection and Union Module for Fine-Grained Visual Classification[c]. IEEE International Conference on Multimedia and Expo (ICME), 2020.

Abstract: One of the predominant methods of previous works on FineGrained Visual Classification (FGVC) is to localize discriminative parts by auxiliary networks and extract part-based finegrained features for classification. In this paper, we propose a simple yet effective approach by introducing an intersection and union module (IU-Module), which groups some feature channels together to capture a set of more discriminative features, along with sharing parts of interest within the group and adding a differentiation loss to reduce the similarity among those grouped feature channels. Without adding any new learnable parameters, our approach imposes two straightforward operations, channel intersection operation (CI) and channel union operation (CU), on convolutional neural network and achieves competitive results compared with stateof-the-art. Experimental results on three publicly available FGVC datasets (CUB-200-2011, Stanford Cars, and FGVCAircraft) show the effectiveness of our approach. Ablation studies and visualizations are provided to further evaluate our approach.

Download paper here