RRGCCAN: Re-Ranking via Graph Convolution Channel Attention Network for Person Re-Identification

Abstract

The classical person re-identification methods are mostly focused on employing discriminative features amongst which the distance is measured on Euclidean space, while the effort of re-ranking is constrained as the lack of the utilization of quality context representation in embedding set. In this paper, we incorporate graph models on feature subsets resorting to the initial ranking by adopting the integration of the attention mechanism into graph convolution network. On the one hand, the context information regarding embedding pairs is considered to compute feature group similarity through the aggregation operation by using graph convolution networks. On the other hand, we adopt a channel attention mechanism to enhance the contribution of relevant feature channels, further strengthening the ability of similarity pulling and dissimilarity pushing of the overall network. Experimental study shows that the proposed network structure is superior to the state-of-the-art deep neural networks on three very challenging datasets that are popular in examining person re-identification techniques.

Software engineering methodologies rely on version control systems such as git to store source code artifacts and manage changes to the codebase. Pull requests include chunks of source code, history of changes, log messages around a proposed change of the mainstream codebase, and much discussion on whether to integrate such changes or not. A better understanding of what contributes to a pull request fate and latency will allow us to build predictive models of what is going to happen and when. Several factors can influence the acceptance of pull requests, many of which are related to the individual aspects of software developers. In this study, we aim to understand how the affect (e.g., sentiment, discrete emotions, and valence-arousal-dominance dimensions) expressed in the discussion of pull request issues influence the acceptance of pull requests. We conducted a mining study of large git software repositories and analyzed more than 150,000 issues with more than 1,000,000 comments in them. We built a model to understand whether the affect and the politeness have an impact on the chance of issues and pull requests to be merged-i.e., the code which fixes the issue is integrated in the codebase. We built two logistic classifiers, one without affect metrics and one with them. By comparing the two classifiers, we show that the affect metrics improve the prediction performance. Our results show that valence (expressed in comments received and posted by a reporter) and joy expressed in the comments written by a reporter are linked to a higher likelihood of issues to be merged. On the contrary, sadness, anger, and arousal expressed in the comments written by a reporter, and anger, arousal, and dominance expressed in the comments received by a reporter, are linked to a lower likelihood of a pull request to be merged.