Shulou(Shulou.com)06/01 Report--

This article introduces you to use the regional attention network to improve R-MAC method is what, the content is very detailed, interested friends can refer to, hope to be helpful to you.

In recent years, many methods of effective image retrieval using convolution neural network (CNN) focus on feature aggregation rather than feature embedding, because convolution features have been found to have reasonable discrimination. Nevertheless, we find that the well-known region-based feature aggregation method R-MAC for image retrieval is still affected by background clutter and region importance. In this work, we solve these problems through a simple and effective context-aware Regional attention Network (context-aware regional attention network), which measures the attention score of a region according to global attention. Experiments on widely used image retrieval datasets show that our method not only significantly improves the R-MAC baseline method, but also reaches a new state-of-the-art height in the "pre-trained single-pass" category method. In addition, we have proved that our method shows a higher accuracy improvement than the previous method when used in conjunction with the query expansion (query expansion) method. These results can be attributed to our new regional focus network integrated with R-MAC.

The figure above shows two challenging examples of background and clutter in Oxford5k. In each example, there is a query on the left and its corresponding positive image on the right, marking the first five areas of interest of our region in the positive image as a red box.

The above figure is the network architecture of this paper. The original R-MAC method treats each image region for feature aggregation equally, while the author introduces Regional attention module to generate different region weights according to the principle of image significance detection, and then R-MAC and regional attention weights are weighted to aggregate to get the final feature vector.

Significant performance improvements have been achieved on multiple datasets, as shown in the following figure:

On the use of regional attention network to improve the R-MAC method is shared here, I hope the above content can be of some help to you, can learn more knowledge. If you think the article is good, you can share it for more people to see.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.