Core Concepts
A novel part-attention based model (PAB-ReID) is proposed to effectively address the challenges in occluded person re-identification by leveraging human parsing labels to generate accurate part attention maps, a fine-grained feature focuser to suppress background interference, and a part triplet loss to learn robust local features.
Abstract
The paper proposes a part-attention based model (PAB-ReID) to tackle the challenges in occluded person re-identification.
Key highlights:
- The part attention block utilizes human parsing labels to guide the generation of accurate part attention maps, providing more precise feature extraction regions for different body parts.
- The fine-grained feature focuser applies the part attention maps to the deep features extracted by the backbone, filtering irrelevant background information and generating fine-grained body part features.
- The part triplet loss is designed to supervise the learning of body part features, enhancing the robustness of the model to similar part appearances.
The authors conduct extensive experiments on specialized occlusion and regular re-identification datasets, demonstrating that PAB-ReID outperforms existing state-of-the-art methods.
Stats
The paper does not provide any specific numerical data or statistics to support the key logics.
Quotes
The paper does not contain any striking quotes supporting the key logics.