Dual-network Multi-attention Collaborative Classification Based on Fine-grained Vision

Fine-grained visual classification aims to identify images that belong to multiple subcategories within the same category. Most existing methods use a single network to extract image features or learn fine-grained features by locating and scaling key regions. Due to the limited number of components, this can ignore valuable clues or cause performance degradation. This paper proposes an efficient method to solve this problem: dual networks encourage the model to learn as many global features as possible. Considering the importance of attention, we utilize the attention generated by the gated interaction mechanism between different channels or the same channel to focus on key feature regions in the global feature map. Discover more discriminative feature maps. The gating vector in the gating mechanism is generated by a multilayer perceptron (MLP) and exploits its interaction with the original feature channel to generate attention. Similarly, bilinear is used to calculate the relationship between the locations and channels of feature maps to find common attention regions between them. The proposed model performs well on three datasets: CUB-200-2011, Stanford car, and FGVC aircraft.