British Machine Vision Conference (BMVC) 2018

Regional Attention Based Deep Feature for Image Retrieval Regional Attention Based Deep Feature for Image Retrieval

by Jaeyoon Kim and Sung-Eui Yoon

Korea Advanced Institute of Science and Technology (KAIST)

This figure shows two challenging examples with backgrounds and clutters in Oxford5k. In each example, the left is a query, while the right is its corresponding positive image. We mark top five attentive regions of our regional attention network in the positive images as red boxes.

This shows our overall encoding sequence of computing feature vector from a given image, based on the R-MAC module and our novel regional attention module.


Many approaches using Convolutional Neural Network (CNN) for efficient image retrieval have concentrated on feature aggregation rather than feature embedding over recent years, since convolutional features have been found to be reasonably discriminative. Nonetheless, we found that a well-known region-based feature aggregation method, R-MAC, for image retrieval is suffered from the background clutter and varying importance of regions. In this work, we tackle these problems with a simple and effective, context-aware regional attention network that weights an attentive score of a region considering global attentiveness. We conduct various experiments on well-known retrieval datasets, and confirm that our method does not only improve the R-MAC baseline significantly, but also present new state-of-the-art results in the category of ``pre-trained single-passĄŻĄŻ. Furthermore, we show that our method shows higher accuracy improvement combined over prior methods, when combined with the query expansion method. These results are attributed by our novel regional-attention network integrated with R-MAC.


Paper (author preprint), Poster
Source code: Github page, ZIP file

Weights: TAR file

  author  = {Jaeyoon Kim and Sung-Eui Yoon},
  title   = {Regional Attention Based Deep Feature for Image Retrieval},
  booktitle = {Proc. British Machine Vision Conference (BMVC 2018)},
  address = {Newcastle, England},
  year = {2018},
  pages = {},
  volume  = {},