Consistent attentive dual branch network for person re-identification

Abstract Several recent person re-identification methods are focusing on learning discriminative representations by designing efficient metric loss functions. Other approaches design part based architectures to compute an informative descriptor local features from semantically coherent parts. Few efforts learn the relationship between distant similar regions and parts adjusting them their most feasible positions with help of soft attention. However, they focus calibrating ignore noise (blur) free distinct feature as datasets contain degraded images. To tackle these issues, we propose a novel Consistent Attention Dual Branch Network (CadNet) that has ability model long-range dependencies (correlations) channels well maps. We adopt multiple classifiers trained global for unique representation person. Correlation consistently computed using channel attention mechanism make learned distict noisy blurry data. Feature correlations interpret similarities in images self mechanism. The proposed CadNet significantly enhances performance respect baseline benchmarks.