Frequency-oriented adaptive real-time object detector for cluttered traffic scenes.

Frequency-Oriented Adaptive Detector achieves up to 90.9% mAP on KITTI with only 8.77 million parameters, outperforming current state-of-the-art traffic object detectors.

• Why it matters: Efficient and accurate traffic object detection is critical for autonomous driving, but balancing these needs on in-vehicle platforms remains a significant challenge due to computational constraints.
• What they did: The approach combines high- and low-frequency information through Frequency Dynamic Convolution and an Adaptive Frequency-Oriented Fusion framework, enabling multi-scale feature enhancement with low overhead.
• The result: This method improves detection accuracy across multiple datasets, enabling real-time, resource-efficient traffic scene understanding that supports safer autonomous driving systems.