QYOLO: Quantum-Inspired Lightweight Object Detection
A new object detection framework called QYOLO achieves architectural compression by replacing deep C2f bottleneck modules with a quantum-inspired QMixBlock. The method uses sinusoidal mixing with shared parameters across backbone stages P4/16 (512 channels) and P5/32 (1024 channels), reducing parameters while maintaining performance. The neck and detection head remain classical.
Key facts
- QYOLO is a quantum-inspired channel mixing framework for object detection.
- It replaces the two deepest C2f bottleneck modules at P4/16 and P5/32 with a QMixBlock.
- The QMixBlock uses sinusoidal mixing with shared learnable parameters across stages.
- The neck and detection head are fully classical and unchanged.
- The approach reduces computational overhead from quadratic scaling with channel width.
- The paper is available on arXiv with ID 2604.26435.
- It targets lightweight object detection for real-time visual perception.
- Single stage detectors are the dominant solution for real-time perception.
Entities
Institutions
- arXiv