Graph Attention Transformer Network for Multi-label Image Classification

Multi-label classification aims to recognize multiple objects or attributes from images. The key solving this issue relies on effectively characterizing the inter-label correlations dependencies, which bring prevailing graph neural network. However, current methods often use co-occurrence probability of labels based training set as adjacency matrix model correlation, is greatly limited by dataset and affects model’s generalization ability. This article proposes a Graph Attention Transformer Network, general framework for multi-label image mining rich effective label correlation. First, we cosine similarity value pre-trained word embedding initial correlation matrix, can represent richer semantic information than one. Subsequently, propose attention transformer layer transfer adapt domain. Our extensive experiments have demonstrated that our proposed achieve highly competitive performance three datasets.