Manuscript received October 27, 2025; revised November 25, 2025; accepted December 16, 2025; published July 16, 2026.

Abstract—This study presents a benchmarking framework for multilabel semantic segmentation of mango surface defects. It tackles the real-world issue of overlapping and co-occurring symptoms in agricultural images. The curated dataset includes full fruit images and detailed patches with multi-class annotations. This setup captures the biological complexity of natural crop conditions. Three existing segmentation architectures, DeepLabV3+, HRNet-MultilabelSeg, and You Only Look Once with Vision Transformer (YOLOViT)-MultilabelSeg, are compared using a unified evaluation method. The framework uses standardized metrics, including mean Intersection-over-Union (mIoU), F1-Score, precision, recall, and inference speed. DeepLabV3+ achieved the highest accuracy. Meanwhile, HRNet-MultilabelSeg provided real-time performance at 526 Frames Per Second (FPS) while maintaining competitive segmentation quality. This contribution offers a clear and reproducible benchmark, along with a practical roadmap for public, cross-domain validation and unified training goals.

Keywords— deep learning in agriculture, multilabel semantic segmentation, pixel-level defect annotation, smart precision farming, DeepLabV3+, HRNet-MultilabelSeg, You Only Look Once with Vision Transformer (YOLOViT)-MultilabelSeg

Cite: Maria Jeseca C. Baculo, Conrado Ruiz Jr., and Oya Aran, "A Benchmarking Framework for Fine-grained Multilabel Semantic Segmentation of Mango Fruit Surface Defects," Journal of Image and Graphics, Vol. 14, No. 4, pp. 532-542, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).