Ask or search anything...

Researchers from the University of Ottawa and Sensor Cortek Inc. introduce mRadNet, a compact and efficient MetaFormer-based model for radar object detection in ADAS. This model achieves an Average Precision of 86.72% and Average Recall of 91.18% on the CRUW dataset, setting a new benchmark for accuracy while significantly reducing computational footprint to 4.93 million parameters and 32.79 GFLOPs.

T-FFTRadNet introduces an object detection framework for radar that processes raw ADC signals using learnable transformation layers and hierarchical Swin Vision Transformers. This approach significantly reduces computational overhead and inference time, achieving competitive performance and improved recall, particularly for low-definition radar.