[论文解读] The KFIoU Loss for Rotated Object Detection

Xue Yang, Yue Zhou+6 more|arXiv (Cornell University)|Jan 29, 2022

Advanced Neural Network Applications被引用 88

一句话总结

本文提出 KFIoU 损失，一种易于实现、完全可微的基于高斯乘积的近似 SkewIoU 的方法，用于旋转目标检测，具有 2-D 和 3-D 扩展，在航空、文本和人脸数据集上显示出强大性能。

ABSTRACT

Differing from the well-developed horizontal object detection area whereby the computing-friendly IoU based loss is readily adopted and well fits with the detection metrics. In contrast, rotation detectors often involve a more complicated loss based on SkewIoU which is unfriendly to gradient-based training. In this paper, we propose an effective approximate SkewIoU loss based on Gaussian modeling and Gaussian product, which mainly consists of two items. The first term is a scale-insensitive center point loss, which is used to quickly narrow the distance between the center points of the two bounding boxes. In the distance-independent second term, the product of the Gaussian distributions is adopted to inherently mimic the mechanism of SkewIoU by its definition, and show its alignment with the SkewIoU loss at trend-level within a certain distance (i.e. within 9 pixels). This is in contrast to recent Gaussian modeling based rotation detectors e.g. GWD loss and KLD loss that involve a human-specified distribution distance metric which require additional hyperparameter tuning that vary across datasets and detectors. The resulting new loss called KFIoU loss is easier to implement and works better compared with exact SkewIoU loss, thanks to its full differentiability and ability to handle the non-overlapping cases. We further extend our technique to the 3-D case which also suffers from the same issues as 2-D. Extensive results on various public datasets (2-D/3-D, aerial/text/face images) with different base detectors show the effectiveness of our approach.

研究动机与目标

• 推动并解决旋转目标检测中旋转 SkewIoU 指标与回归损失之间的不对齐问题。
• 提出基于高斯建模的可微分、无超参数的损失来近似 SkewIoU。
• 将方法从 2-D 扩展到 3-D 旋转检测，并在多样数据集和检测器上进行验证。

提出的方法

• 通过旋转和特征分解（R, Λ）将旋转边界框转换为高斯分布。
• 使用一个与尺度无关的中心点损失来缩小高斯之间的中心距离。
• 通过高斯分布的乘积计算重叠，并推导出以重叠为基础的归一化损失的 KFIoU。
• 将回归损失表述为 L_reg = L_c + L_kf，其中 L_kf = exp(1 - KFIoU) - 1。
• 提供两种中心点损失 L_c 的选项（标准基于 L_n 的或基于 KLD 的中心项）。
• 保持端到端可微性并适用于不重叠情况，具有 2-D 和 3-D 的扩展。

实验结果

研究问题

• RQ1 是否存在一个完全可微的基于高斯乘积的损失，在不进行超参数调整的情况下即可近似旋转框的 SkewIoU？
• RQ2 与 GWD、KLD 和普通 SkewIoU 相比，提出的 KFIoU 损失是否在 2-D 和 3-D 任务以及多样数据集上提升旋转检测性能？
• RQ3 在不同距离和宽高比下，KFIoU 与 SkewIoU 的趋势一致性与其他基于高斯的损失（GWD、KLD）相比如何？

主要发现

• KFIoU 损失易于使用标准 DL 运算实现，并能有效处理不重叠情况。
• KFIoU 与 SkewIoU 在趋势层面对齐优于 GWD 和 KLD，并降低对超参数的敏感性。
• 在多个数据集（航空、场景文本、人脸）和检测器上，KFIoU 的表现优于最优调参的高斯基方法和普通 SkewIoU 变体。
• 将高斯建模扩展到 3-D 旋转检测，在 KITTI BEV/3-D 指标上对基线取得显著改进。