Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation for Occluded Person Re-Identification
Occluded person re-identification aims to retrieve holistic images of a given identity based on occluded person images. Most existing approaches primarily focus on aligning visible body parts using prior information, applying occlusion augmentation to predefined regions, or complementing the missing semantics of occluded body parts with the assistance of holistic images. Nevertheless, they struggle to generalize across diverse occlusion scenarios that are absent from the training data and often overlook the pervasive issue of feature contamination caused by holistic images. In this work, we propose a novel Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation (OGFR) to address these two issues simultaneously. OGFR adopts a teacher-student distillation architecture that effectively incorporates diverse occlusion patterns into feature representation while transferring the purified discriminative holistic knowledge from the holistic to the occluded branch through reinforced knowledge distillation. Specifically, an Occlusion-Aware Vision Transformer is designed to leverage learnable occlusion pattern embeddings to explicitly model such diverse occlusion types, thereby guiding occlusion-aware robust feature representation. Moreover, we devise a Feature Erasing and Purification Module within the holistic branch, in which an agent is employed to identify low-quality patch tokens of holistic images that contain noisy negative information via deep reinforcement learning, and substitute these patch tokens with learnable embedding tokens to avoid feature contamination and further excavate identity-related discriminative clues. Afterward, with the assistance of knowledge distillation, the student branch effectively absorbs the purified holistic knowledge to precisely learn robust representation regardless of the interference of occlusions. {Extensive experiments validate OGFR, on Occluded-Duke it achieves 76.6% Rank-1 and 64.7% mAP, outperforming the closest Transformer-based method by +3.3% Rank-1 and +2.4% mAP, with consistent gains on other benchmarks.
Paper
Full text
Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation for Occluded Person Re-Identification
Semantic Scholar · Computer Science · 2025
Abstract
Occluded person re-identification aims to retrieve holistic images of a given identity based on occluded person images. Most existing approaches primarily focus on aligning visible body parts using prior information, applying occlusion augmentation to predefined regions, or complementing the missing semantics of occluded body parts with the assistance of holistic images. Nevertheless, they struggle to generalize across diverse occlusion scenarios that are absent from the training data and often overlook the pervasive issue of feature contamination caused by holistic images. In this work, we propose a novel Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation (OGFR) to address these two issues simultaneously. OGFR adopts a teacher-student distillation architecture that effectively incorporates diverse occlusion patterns into feature representation while transferring the purified discriminative holistic knowledge from the holistic to the occluded branch through reinforced knowledge distillation. Specifically, an Occlusion-Aware Vision Transformer is designed to leverage learnable occlusion pattern embeddings to explicitly model such diverse occlusion types, thereby guiding occlusion-aware robust feature representation. Moreover, we devise a Feature Erasing and Purification Module within the holistic branch, in which an agent is employed to identify low-quality patch tokens of holistic images that contain noisy negative information via deep reinforcement learning, and substitute these patch tokens with learnable embedding tokens to avoid feature contamination and further excavate identity-related discriminative clues. Afterward, with the assistance of knowledge distillation, the student branch effectively absorbs the purified holistic knowledge to precisely learn robust representation regardless of the interference of occlusions. {Extensive experiments validate OGFR, on Occluded-Duke it achieves 76.6% Rank-1 and 64.7% mAP, outperforming the closest Transformer-based method by +3.3% Rank-1 and +2.4% mAP, with consistent gains on other benchmarks.