3DSMCOS: A 3D Model-Based Synthetic Data Pipeline for Military Camouflaged Object Segmentation with Distractor-Augmented Realism

The detection and segmentation of military camouflaged objects remain challenging due to complex camouflage patterns, environmental variability, and the limited availability of annotated training data. To address these issues, we propose 3DSMCOS, a novel pipeline for generating high-quality synthetic datasets using 3D models. This pipeline integrates distractor objects into the scene to enhance realism and guide model attention toward military camouflaged targets. In addition, we introduce SAMMCOS, an automated segmentation pipeline that converts existing bounding box annotations into accurate pixel-level masks, providing complementary training data. We evaluate our approach using several state-of-the-art camouflaged object detection models, including SINetV2, MRR-Net, MFFN, DGNet, and ZoomNet, and demonstrate that it significantly improves both segmentation and detection performance. These results highlight the potential of domain-specific synthetic data and automated labeling in advancing robust and generalizable military camouflage object detection systems. Datasets and additional resources can be found at https://github.com/hangtruong85/MHCD-Seg.