Image Tagging with Deep Learning: Fine‐Grained Visual Analysis
Jianlong Fu, Tao Mei · 2019
This chapter introduces the techniques and applications of deep learning frameworks on fine-grained image tagging that is recognizing image sub-categories or visual sentiment analysis. It provides more comparison between different deep learning models and deeper analysis with visualization results over widely used fine-grained image tagging datasets. The chapter then introduces the recurrent attention convolutional neural network (RA-CNN) for fine-grained image tagging. Standford Dogs is another widely used dataset for fine-grained image recognition. The chapter considers the visual sentiment analysis as a binary prediction problem which classifies an image as positive or negative from its visual content. It proposes another designed deep neural network architecture that is a deep coupled adjective and noun network (DCAN). Adjectives are usually related to descriptiveness, while nouns represent the objectiveness of an image. SentiBank is widely used and contains about half a million images from Flickr using the designed adjective noun pairs as queries.