A Self-Supervised Tree-Structured Framework for Fine-Grained Classification

In computer vision, fine-grained classification has become an important issue in recognizing objects with slight visual differences. Usually, it is challenging to generate good performance when solving problems using traditional convolutional neural networks. To improve the accuracy and training time of networks problems, this paper proposes a tree-structured framework by eliminating effect differences between clusters. The contributions proposed method include following three aspects: (1) self-supervised that automatically creates tree, need for manual labeling; (2) machine-learning matcher which determines cluster item belongs, minimizing impact inter-cluster variations on classification; (3) pruning criterion filters classifier, retaining only models superior performance. experimental evaluation demonstrates its effectiveness reducing improving across various datasets comparison conventional network models. Specifically, CUB 200 2011, FGVC aircraft, Stanford car datasets, achieves reduction 32.91%, 35.87%, 14.48%, improves 1.17%, 2.01%, 0.59%, respectively.