Hybrid-attention guided network with multiple resolution features for person re-identification

Extracting effective and discriminative features is highly important for addressing the challenges of person re-identification (re-ID). At present, deep convolutional neural networks typically use high-level to identify pedestrians. However, some essential spatial information contained in low-level will be lost during learning feature. Most existing re-ID methods rely primarily on hand-crafted bounding boxes , which images are precisely aligned. these approaches unrealistic practical applications due inaccuracy automatic detection algorithms . To address problems, we propose a hybrid-attention guided network with multiple resolution re-ID. First, construct multi-resolution fusion strategy ensure that can spatially aligned feature fusion, while at same time ensuring discriminability after fusion. We then introduce an attention mechanism multi-granularity operation reduce impact irregular by gently expanding size maps. In addition, new multi-pool extractor designed obtain different types using two pools, enabling representation capability further improved. Extensive experiments demonstrate superiority our approach. Our code available https://github.com/libraflower/MutipleFeature-for-PRID.