RiO-DETR: Real-Time Oriented Object Detection with DETR, Achieving 2.7ms Latency

机器之心 · wechat · 2026-08-16

Researchers from USTC, Huawei, and Tsinghua propose RiO-DETR, pushing end-to-end oriented object detection into real-time. On DOTA-1.0 single-scale, the lightest model has 4.0M params, 2.7ms latency, and 78.4 AP50; the strongest achieves 81.8 AP50 at 29.9ms. Accepted as Oral at ECCV 2026. Key designs include content-driven angle estimation, rotation-rectified orthogonal attention, decoupled periodic refinement, and oriented dense O2O training.

Original post →

More from Research

Research channel →