ID-aware Quality for Set-based Person Re-identification
This work improves person re-identification accuracy for surveillance and security applications, though it is incremental as it builds on existing SReID models.
The paper tackles the problem of set-based person re-identification (SReID) by addressing noise and overfitting issues with a novel ID-aware quality method, achieving state-of-the-art performance on four datasets.
Set-based person re-identification (SReID) is a matching problem that aims to verify whether two sets are of the same identity (ID). Existing SReID models typically generate a feature representation per image and aggregate them to represent the set as a single embedding. However, they can easily be perturbed by noises--perceptually/semantically low quality images--which are inevitable due to imperfect tracking/detection systems, or overfit to trivial images. In this work, we present a novel and simple solution to this problem based on ID-aware quality that measures the perceptual and semantic quality of images guided by their ID information. Specifically, we propose an ID-aware Embedding that consists of two key components: (1) Feature learning attention that aims to learn robust image embeddings by focusing on 'medium' hard images. This way it can prevent overfitting to trivial images, and alleviate the influence of outliers. (2) Feature fusion attention is to fuse image embeddings in the set to obtain the set-level embedding. It ignores noisy information and pays more attention to discriminative images to aggregate more discriminative information. Experimental results on four datasets show that our method outperforms state-of-the-art approaches despite the simplicity of our approach.