通过不确定性门控检索提升街景语义分割的鲁棒性
Uncertainty-Gated Region-Level Retrieval for Robust Semantic Segmentation
- 基于区域不确定性的门控机制，只对高不确定区域调用外部检索
- 在域偏移下mIoU提升11.3%，仅需检索12.5%区域
- 适合自动驾驶等需实时高精度分割的场景
室外街景语义分割在自动驾驶、移动机器人及视障人士辅助技术中至关重要。准确区分道路、人行道、车辆和行人等关键表面与物体，是保障安全、降低风险的关键。分割模型需在不同环境、光照、天气条件及传感器噪声下保持鲁棒性，并支持实时处理。本文提出一种区域级不确定性门控检索机制，在域偏移条件下提升分割精度与校准能力。最佳方法在平均交并比（mIoU）上提升11.3%，同时将检索成本降低87.5%，仅需对12.5%的区域进行检索，而基准方案始终开启需100%检索。
原文摘要 · Abstract (English)
Semantic segmentation of outdoor street scenes plays a key role in applications such as autonomous driving, mobile robotics, and assistive technology for visually-impaired pedestrians. For these applications, accurately distinguishing between key surfaces and objects such as roads, sidewalks, vehicles, and pedestrians is essential for maintaining safety and minimizing risks. Semantic segmentation must be robust to different environments, lighting and weather conditions, and sensor noise, while being performed in real-time. We propose a region-level, uncertainty-gated retrieval mechanism that improves segmentation accuracy and calibration under domain shift. Our best method achieves an 11.3% increase in mean intersection-over-union while reducing retrieval cost by 87.5%, retrieving for only 12.5% of regions compared to 100% for always-on baseline.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。