Transfer Learning Approach to Fine-Grained Image Classification
Valentin A. Golodov, Mariya S. Dubrovina, Anastasiya S. Paziy · 2019 International Russian Automation Conference (RusAutoCon) · 2019
This paper deals with fine-grained image classification in which instances from different classes share common parts but have wide variation in shape and appearance. Introduction gives a short review of existing works on this topic such as weakly supervised learning, discriminative localization network. Various approaches to the solution were considered and the method of the transfer of learning for the fine-grained image classification is considered as one the most promising methods. Convolutional neural networks are used. The approach demonstrates its effectiveness and allowed obtaining high accuracy of recognition on the dog breed recognition problem. Next step is to add more breeds and improve the recognition result.