MQADet introduces a universal paradigm that enhances existing open-vocabulary object detection systems by leveraging multimodal question answering capabilities.

• Addresses visual-textual misalignment and long-tailed category imbalances in current systems
• Serves as a plug-and-play solution compatible with existing open-vocabulary detectors
• Improves detection performance for previously unseen objects
• Particularly valuable for security applications like surveillance and threat detection where identifying unknown objects is critical

Security Impact: By enabling systems to accurately identify objects beyond their training categories, MQADet significantly enhances security monitoring capabilities without requiring extensive retraining.

MQADet: A Plug-and-Play Paradigm for Enhancing Open-Vocabulary Object Detection via Multimodal Question Answering