High-Quality Pseudo-Label Generation Based on Visual Prompt Assisted Cloud Model Update
Xinrun Xu, Qiuhong Zhang, Jianwen Yang, Zhanbiao Lian, Jin Yan, Zhiming Ding, Shan Jiang
TL;DR
CA-HQP tackles the problem of domain drift in cloud-side pseudo-label generation for cloud–edge object detection under evolving traffic conditions. It introduces a Learnable Visual Prompt Generator (VPG) to enable parameter-efficient cloud adaptation and couples global Domain Query Feature Alignment (DQFA) with instance-aware Temporal Instance-Aware Feature Embedding Alignment (TIAFA) to bridge source and target domains. The optimization combines a standard detection loss with adversarial domain losses into $\ L_{all}=\mathcal{L}_{det} - (\mathcal{L}_{adv}^{DQFA} + \mathcal{L}_{adv}^{TIAFA})$, facilitating robust, data-efficient adaptation. Experiments on the Bellevue traffic dataset show consistent improvements in pseudo-label quality and edge-model mAP across multiple baselines, validating CA-HQP's effectiveness for reliable cloud–edge collaboration in dynamic environments.
Abstract
Generating high-quality pseudo-labels on the cloud is crucial for cloud-edge object detection, especially in dynamic traffic monitoring where data distributions evolve. Existing methods often assume reliable cloud models, neglecting potential errors or struggling with complex distribution shifts. This paper proposes Cloud-Adaptive High-Quality Pseudo-label generation (CA-HQP), addressing these limitations by incorporating a learnable Visual Prompt Generator (VPG) and dual feature alignment into cloud model updates. The VPG enables parameter-efficient adaptation by injecting visual prompts, enhancing flexibility without extensive fine-tuning. CA-HQP mitigates domain discrepancies via two feature alignment techniques: global Domain Query Feature Alignment (DQFA) capturing scene-level shifts, and fine-grained Temporal Instance-Aware Feature Embedding Alignment (TIAFA) addressing instance variations. Experiments on the Bellevue traffic dataset demonstrate that CA-HQP significantly improves pseudo-label quality compared to existing methods, leading to notable performance gains for the edge model and showcasing CA-HQP's adaptation effectiveness. Ablation studies validate each component (DQFA, TIAFA, VPG) and the synergistic effect of combined alignment strategies, highlighting the importance of adaptive cloud updates and domain adaptation for robust object detection in evolving scenarios. CA-HQP provides a promising solution for enhancing cloud-edge object detection systems in real-world applications.