Deep convolutional hamming ranking network for large scale image retrieval
Shi Zhong, Kai Li, Rui Feng · 2014
In this paper we address the problem of large image retrieval from millions of images. Recently, deep convolutional neural network has demonstrated superior performance in a number of computer vision applications. We propose to adapt the existing architecture targeted towards image classification to directly learn features for efficient image retrieval. We extend the Weighted Approximate Rank Pairwise(WARP) loss to the Hamming space for learning binary features. The features learned with the ranking loss achieve higher accuracy. Extensive experiments demonstrate competitive performance on five public benchmark datasets UKbench, Holidays, Oxford Buildings, Paris Buildings and San Francisco Landmarks.