Reduced Focal Loss: 1st Place Solution to xView object detection in Satellite Imagery

TL;DR

This paper presents a novel Reduced Focal Loss function that effectively handles class imbalance in satellite imagery object detection, leading to winning the DIUx xView 2018 challenge.

Contribution

Introduction of a new Reduced Focal Loss function specifically designed for imbalanced satellite imagery datasets.

Findings

Achieved 1st place in the xView detection challenge

Improved detection performance on highly imbalanced classes

Demonstrated effectiveness of the new loss function in real-world competition

Abstract

This paper describes our approach to the DIUx xView 2018 Detection Challenge [1]. This challenge focuses on a new satellite imagery dataset. The dataset contains 60 object classes that are highly imbalanced. Due to the imbalanced nature of the dataset, the training process becomes significantly more challenging. To address this problem, we introduce a novel Reduced Focal Loss function, which brought us 1st place in the DIUx xView 2018 Detection Challenge.