RAVEN：面向高效逐 chirp 目标检测与分割的雷达自适应视觉编码器
RAVEN: Radar Adaptive Vision Encoders for Efficient Chirp-wise Object Detection and Segmentation
对做车载雷达感知的人，这篇给出了一个可复用的思路：用可学习的天线混合模块把虚拟 MIMO 阵列压成紧凑特征，再配合标定过的提前退出规则做逐 chirp 推理，从而在雷达帧内只用部分 chirp 就出结果；这套「按需计算」的设定可直接迁移到其他时序传感器的高效感知骨干上。
原文依据 performing per-chirp inference with a calibrated early-exit rule, so the model reaches a decision using only a subset of chirps
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出 RAVEN，一种用于处理调频连续波（FMCW）雷达数据的深度学习架构，其设计目标是实现高计算效率。RAVEN 通过在独立的接收机状态空间编码器（SSM）上使用可学习的天线混合模块，将虚拟 MIMO 阵列压缩为一组紧凑的学习特征，并借助标定过的提前退出规则执行逐 chirp 推理，从而降低计算量；由此模型只需使用雷达帧中的一部分 chirp 即可做出决策。这些设计选择相比传统的基于帧的雷达骨干网络，可实现最高 170 倍的计算量降低和 4 倍的端到端延迟降低，同时在车载雷达数据集上取得最优的检测与鸟瞰图自由空间分割性能。
英文原文摘要 767 字符
We introduce RAVEN, a deep learning architecture for processing frequency-modulated continuous-wave (FMCW) radar data that is designed for high computational efficiency. RAVEN reduces computation by using a learnable antenna mixer module on independent receiver state space encoders (SSM) to compress the virtual MIMO array into a compact set of learned features and by performing per-chirp inference with a calibrated early-exit rule, so the model reaches a decision using only a subset of chirps in a radar frame. These design choices yield up to 170x lower computation and 4x lower end-to-end latency than conventional frame-based radar backbones, while achieving state-of-the-art detection and BEV free-space segmentation performance on automotive radar datasets.