Feature Completion for Occluded Person Re-Identification.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Person re-identification (reID) is crucial in computer vision but struggles with occluded scenes.
• Existing reID methods often discard occluded regions, leading to performance degradation.

Purpose of the Study:

• To propose an occlusion-robust block, Region Feature Completion (RFC), for improving person re-identification.
• To develop a method that recovers semantic information from occluded regions instead of discarding them.

Main Methods:

• Introduced a Spatial RFC (SRFC) module using long-range spatial contexts to predict features for occluded regions via an encoder/decoder architecture.
• Developed a Temporal RFC (TRFC) module to refine SRFC predictions using long-term temporal contexts.
• Integrated the RFC block into Convolutional Neural Networks (CNNs) to create the RFCnet model, designed for end-to-end training.

Main Results:

• The proposed RFC block successfully recovers semantics of occluded regions in the feature space.
• RFCnet significantly outperforms existing methods on occluded reID benchmarks.
• The method maintains superior performance on holistic reID datasets as well.

Conclusions:

• The RFC block offers an effective solution for occlusion challenges in person re-identification.
• RFCnet demonstrates robustness and high performance across both occluded and holistic reID scenarios.
• The lightweight and plug-and-play nature of RFC facilitates its integration into existing reID systems.