Scale-Aware Relay and Scale-Adaptive Loss for Tiny Object Detection in Aerial Images

TL;DR

This work tackles tiny object detection in aerial imagery by introducing Scale-Aware Relay Layer (SARL) and Scale-Adaptive Loss (SAL). SARL adds a cross-scale spatial-channel attention mechanism to enrich and propagate discriminative features across pyramid levels, while SAL dynamically scales IoU-based regression penalties to focus training on small objects. Across AI-TOD, DOTA-v2.0, and VisDrone2019, integrating SARL and SAL yields consistent gains, including up to 5.5% AP improvements on YOLOv5 and YOLOx baselines and 29.0% AP on a noisy AI-TOD-v2.0 dataset, demonstrating improved robustness and generalization. The approach is compatible with both anchor-based and anchor-free detectors, offering a practical, plug-in enhancement for TOD in real-world aerial applications.

Abstract

Recently, despite the remarkable advancements in object detection, modern detectors still struggle to detect tiny objects in aerial images. One key reason is that tiny objects carry limited features that are inevitably degraded or lost during long-distance network propagation. Another is that smaller objects receive disproportionately greater regression penalties than larger ones during training. To tackle these issues, we propose a Scale-Aware Relay Layer (SARL) and a Scale-Adaptive Loss (SAL) for tiny object detection, both of which are seamlessly compatible with the top-performing frameworks. Specifically, SARL employs a cross-scale spatial-channel attention to progressively enrich the meaningful features of each layer and strengthen the cross-layer feature sharing. SAL reshapes the vanilla IoU-based losses so as to dynamically assign lower weights to larger objects. This loss is able to focus training on tiny objects while reducing the influence on large objects. Extensive experiments are conducted on three benchmarks (\textit{i.e.,} AI-TOD, DOTA-v2.0 and VisDrone2019), and the results demonstrate that the proposed method boosts the generalization ability by 5.5\% Average Precision (AP) when embedded in YOLOv5 (anchor-based) and YOLOx (anchor-free) baselines. Moreover, it also promotes the robust performance with 29.0\% AP on the real-world noisy dataset (\textit{i.e.,} AI-TOD-v2.0).

Figures (4)

• Figure 1: Schematic illustration of the comparison between the prevailing framework and our framework. (a) Prevailing Framework. (b) Our Framework.
• Figure 2: Overall framework of our method. $SA^2N$ is used to enhance discriminative features. $SFL$ is used to balance the regression loss.
• Figure 3: Position and overall structure of $SA^2N$.
• Figure 4: Specific structure of $FIM$ and $FEM$.