Focaler-IoU: More Focused Intersection over Union Loss

TL;DR

This work analyzes how the distribution of difficult versus easy samples impacts bounding box regression in object detection and introduces Focaler-IoU, which reconstructs IoU-based losses through linear interval mapping using thresholds $d$ and $u$ to emphasize different regression samples. The method yields $IoU^{\text{focaler}}$ values, enabling losses like $L_{\text{Focaler-GIoU}}$ and $L_{\text{Focaler-SIoU}}$ to focus on hard or easy samples as appropriate. The authors provide a formal analysis of sample difficulty, define the loss transformation, and validate the approach on PASCAL VOC with YOLOv8 and AI-TOD with YOLOv5, achieving consistent accuracy gains over standard IoU-based losses. The results demonstrate the practical utility of adapting regression losses to dataset characteristics, with code made available for reproducibility.

Abstract

Bounding box regression plays a crucial role in the field of object detection, and the positioning accuracy of object detection largely depends on the loss function of bounding box regression. Existing researchs improve regression performance by utilizing the geometric relationship between bounding boxes, while ignoring the impact of difficult and easy sample distribution on bounding box regression. In this article, we analyzed the impact of difficult and easy sample distribution on regression results, and then proposed Focaler-IoU, which can improve detector performance in different detection tasks by focusing on different regression samples. Finally, comparative experiments were conducted using existing advanced detectors and regression methods for different detection tasks, and the detection performance was further improved by using the method proposed in this paper.Code is available at \url{https://github.com/malagoutou/Focaler-IoU}.

Figures (1)

• Figure 1: As shown in the figure: (a) and (b) represent the linear interval mapping curves for difficult and simple samples, respectively.