Similar Feature Extraction Network for Occluded Person Re-identification.

Occluded person re-identification (Re-ID) is a challenging task in real-world scenarios due to the extensive conditions that persons are occluded by various obstacles. Although state-of-the-art methods with additional cues such as pose estimation and segmentation have achieved great success, they did not overcome data bias and the dependency on the accuracy of other detectors. In this paper, we propose a novel similar feature extraction network (SFE-Net) for occluded person Re-ID to address these issues. Firstly, we introduce the adaptive convolution method to separate the features of occluded and non-occluded regions, where local and global features are sufficiently used. We then apply adaptive aggregating parameters to find a better weighting strategy automatically. Finally, the transformer encoder architecture is utilized for generating discriminative features. Extensive experiments show SFE-Net outperforms state-of-the-art methods on both occluded and holistic datasets.