Domain Adaptive Transfer Learning on Visual Attention Aware Data Augmentation for Fine-grained Visual Categorization

Fine-Grained Visual Categorization (FGVC) is a challenging topic in computer\nvision. It is a problem characterized by large intra-class differences and\nsubtle inter-class differences. In this paper, we tackle this problem in a\nweakly supervised manner, where neural network models are getting fed with\nadditional data using a data augmentation technique through a visual attention\nmechanism. We perform domain adaptive knowledge transfer via fine-tuning on our\nbase network model. We perform our experiment on six challenging and commonly\nused FGVC datasets, and we show competitive improvement on accuracies by using\nattention-aware data augmentation techniques with features derived from deep\nlearning model InceptionV3, pre-trained on large scale datasets. Our method\noutperforms competitor methods on multiple FGVC datasets and showed competitive\nresults on other datasets. Experimental studies show that transfer learning\nfrom large scale datasets can be utilized effectively with visual attention\nbased data augmentation, which can obtain state-of-the-art results on several\nFGVC datasets. We present a comprehensive analysis of our experiments. Our\nmethod achieves state-of-the-art results in multiple fine-grained\nclassification datasets including challenging CUB200-2011 bird, Flowers-102,\nand FGVC-Aircrafts datasets.\n