Optimizing Real-Time NIR Image Segmentation: Enhancing Accuracy Through Bilateral Fusion for False Negative Mitigation

초록

False negative (FN) and false positive (FP) are one of the main causes of poor performance of tasks such as image segmentation and detection. To reduce the frequency of occurrence of FN and FP, this paper proposes bilateral fusion to strengthen information exchange within a deep learning model with a multi-branch structure. The proposed bilateral fusion improved the performance of the detail-context branch structure model, which is currently widely used in real-time semantic segmentation tasks, and can be easily applied to other models with similar structures. © 2024 IEEE.

키워드