Progressive Co-Attention Network for Fine-grained Visual Classification

• DOI: 10.48550/arxiv.2101.08527
• Publication Date: 2021-01-01
• Authors: Tian Zhang; Dongliang Chang; Zhanyu Ma; Jun Guo
• Keywords: Discriminative model; Benchmark (surveying); Similarity (geometry); Feature (linguistics); Attention network; Contextual image classification

Fine-grained visual classification aims to recognize images belonging multiple sub-categories within a same category. It is challenging task due the inherently subtle variations among highly-confused categories. Most existing methods only take an individual image as input, which may limit ability of models contrastive clues from different images. In this paper, we propose effective method called progressive co-attention network (PCA-Net) tackle problem. Specifically, calculate channel-wise similarity by encouraging interaction between feature channels same-category pairs capture common discriminative features. Considering that complementary information also crucial for recognition, erase prominent areas enhanced channel force focus on other regions. The proposed model has achieved competitive results three fine-grained benchmark datasets: CUB-200-2011, Stanford Cars, and FGVC Aircraft.