Tackling the Background Bias in Sparse Object Detection via Cropped Windows

TL;DR

This paper introduces a tiling method that reduces background bias and enhances small object detection in UAV imagery without altering existing models, validated across multiple datasets with improved performance and speed.

Contribution

The proposed tiling approach effectively mitigates background bias and allows higher resolution training, significantly boosting detection accuracy in sparse UAV imagery.

Findings

Improved detection performance across three datasets

Faster inference compared to similar methods

Enhanced capability to detect small objects in UAV images

Abstract

Object detection on Unmanned Aerial Vehicles (UAVs) is still a challenging task. The recordings are mostly sparse and contain only small objects. In this work, we propose a simple tiling method that improves the detection capability in the remote sensing case without modifying the model itself. By reducing the background bias and enabling the usage of higher image resolutions during training, our method can improve the performance of models substantially. The procedure was validated on three different data sets and outperformed similar approaches in performance and speed.