A survey of fine-grained image categorization
2018 14th IEEE International Conference on Signal Processing (ICSP), 2018•ieeexplore.ieee.org
It is known that the ability of humans to perform basic-level visual classification (eg birds
versus dogs) develops well before their ability to perform fine-grained visual categorization
(eg distinguish different sub-categories from Hooded Oriole to Scott Oriole). Similarly,
computer vision research follows such trajectory. Basic-level image classification has made
great progress with the help of deep learning models in recent years, while fine-grained
image categorization is still facing a number of challenges and attracting more and more …
versus dogs) develops well before their ability to perform fine-grained visual categorization
(eg distinguish different sub-categories from Hooded Oriole to Scott Oriole). Similarly,
computer vision research follows such trajectory. Basic-level image classification has made
great progress with the help of deep learning models in recent years, while fine-grained
image categorization is still facing a number of challenges and attracting more and more …
It is known that the ability of humans to perform basic-level visual classification (e.g. birds versus dogs) develops well before their ability to perform fine-grained visual categorization (e.g. distinguish different sub-categories from Hooded Oriole to Scott Oriole). Similarly, computer vision research follows such trajectory. Basic-level image classification has made great progress with the help of deep learning models in recent years, while fine-grained image categorization is still facing a number of challenges and attracting more and more attention. In this paper, we review the recent progress in fine-grained image categorization. Starting from its definition, we give a brief introduction to some recent developments in fine-grained image categorization. After that, we elaborate different algorithms from strongly supervised learning and weakly supervised learning, and compare their performances on four publicly available benchmarks. Finally, we provide a brief summary of these methods as well as the potential future research directions, i.e. suggest to explore deeper neural networks and generative adversarial networks.
ieeexplore.ieee.org