Towards Resolving the Challenge of Long-tail Distribution in UAV Images for Object Detection

TL;DR

This paper introduces DSHNet, a novel approach for addressing long-tail class distribution in UAV image object detection, significantly improving tail class performance and achieving state-of-the-art results on UAV datasets.

Contribution

The paper proposes DSHNet, the first method specifically designed to resolve long-tail distribution issues in UAV image object detection, with dual-path components for tail and head classes.

Findings

DSHNet significantly boosts tail class detection performance.

Outperforms existing methods on VisDrone and UAVDT datasets.

Achieves state-of-the-art results with combined image cropping techniques.

Abstract

Existing methods for object detection in UAV images ignored an important challenge - imbalanced class distribution in UAV images - which leads to poor performance on tail classes. We systematically investigate existing solutions to long-tail problems and unveil that re-balancing methods that are effective on natural image datasets cannot be trivially applied to UAV datasets. To this end, we rethink long-tailed object detection in UAV images and propose the Dual Sampler and Head detection Network (DSHNet), which is the first work that aims to resolve long-tail distribution in UAV images. The key components in DSHNet include Class-Biased Samplers (CBS) and Bilateral Box Heads (BBH), which are developed to cope with tail classes and head classes in a dual-path manner. Without bells and whistles, DSHNet significantly boosts the performance of tail classes on different detection frameworks. Moreover, DSHNet significantly outperforms base detectors and generic approaches for long-tail problems on VisDrone and UAVDT datasets. It achieves new state-of-the-art performance when combining with image cropping methods. Code is available at https://github.com/we1pingyu/DSHNet