Fine-Grained Adversarial Semi-Supervised Learning

In this article, we exploit Semi-Supervised Learning ( SSL ) to increase the amount of training data improve performance Fine-Grained Visual Categorization FGVC ). This problem has not been investigated in past spite prohibitive annotation costs that requires. Our approach leverages unlabeled with an adversarial optimization strategy which internal features representation is obtained a second-order pooling model. combination allows one back-propagate information parts, represented by pooling, onto setting. We demonstrate effectiveness combined use conducting experiments on six state-of-the-art fine-grained datasets, include Aircrafts, Stanford Cars, CUB-200-2011, Oxford Flowers, Dogs, and recent iNaturalist-Aves. Experimental results clearly show our proposed method better than only previous examined problem; it also higher classification accuracy respect supervised learning methods compared.