Improved Hard Example Mining Approach for Single Shot Object Detectors
This work addresses imbalanced training sets in object detection for UAV applications, but it is incremental as it adapts and combines existing methods.
The paper tackles performance improvement for hard examples in single-shot object detectors by combining two existing hard example mining approaches (LRM and focal loss) in YOLOv5, resulting in a 3% mAP increase over the original loss function and 1-2% gains over individual methods on the 2021 Anti-UAV Challenge Dataset.
Hard example mining methods generally improve the performance of the object detectors, which suffer from imbalanced training sets. In this work, two existing hard example mining approaches (LRM and focal loss, FL) are adapted and combined in a state-of-the-art real-time object detector, YOLOv5. The effectiveness of the proposed approach for improving the performance on hard examples is extensively evaluated. The proposed method increases mAP by 3% compared to using the original loss function and around 1-2% compared to using the hard-mining methods (LRM or FL) individually on 2021 Anti-UAV Challenge Dataset.