PROJECT TITLE :

Object-Part Attention Model for Fine-Grained Image Classification - 2018

ABSTRACT:

Fine-grained Image Classification is to acknowledge hundreds of subcategories belonging to the identical basic-level category, like 200 subcategories belonging to the bird, which is extremely difficult due to large variance in the same subcategory and little variance among totally different subcategories. Existing strategies usually initial find the objects or elements and then discriminate that subcategory the image belongs to. However, they mainly have two limitations: 1) hoping on object or part annotations that are heavily labor consuming; and 2) ignoring the spatial relationships between the object and its elements as well as among these parts, each of which are significantly helpful for finding discriminative parts. Therefore, this Project proposes the article-half attention model (OPAM) for weakly supervised fine-grained image classification and the main novelties are: one) object-part attention model integrates 2 level attentions: object-level attention localizes objects of pictures, and part-level attention selects discriminative elements of object. Both are jointly utilized to find out multi-view and multi-scale features to boost their mutual promotion; and 2) Object-part spatial constraint model combines two spatial constraints: object spatial constraint ensures selected components highly representative and part spatial constraint eliminates redundancy and enhances discrimination of selected elements. Both are jointly used to take advantage of the subtle and local differences for distinguishing the subcategories. Importantly, neither object nor part annotations are used in our proposed approach, that avoids the significant labor consumption of labeling. Compared with more than ten state-of-the-art ways on four widely-used datasets, our OPAM approach achieves the best performance.