EVIAC

The EVIAC dataset is a large-scale enhanced benchmark for visible\u2013infrared camouflaged object detection, built upon our previous VIAC dataset of 1,500 visible\u2013infrared pairs captured via a dual-spectrum UAV system. To enhance diversity, a style transfer algorithm is applied to generate three additional environments\u2014forest, fog, and snow\u2014yielding 6,000 precisely aligned visible\u2013infrared pairs across four scenes. EVIAC maintains structural and semantic consistency during style transfer, ensuring automatic ground-truth inheritance without extra labeling. It provides rich variability in illumination, object scale, and distribution, and includes annotations for 9 challenge factors (BCO, SCO, MCO, CB, OV, OC, TC, IC, LI). This makes EVIAC a diverse, reliable, and cost-efficient benchmark for studying multimodal perception under complex camouflage conditions.EVIAC数据集是一款面向可见光-红外伪装目标检测的大规模增强型基准数据集，其构建依托于我们此前基于双光谱无人机系统采集的VIAC数据集——该数据集包含1500组可见光-红外图像对。为提升数据集多样性，研究团队采用风格迁移算法生成森林、雾天、雪地三类额外场景，最终得到覆盖四类场景、共计6000组精准对齐的可见光-红外图像对。EVIAC在风格迁移过程中保持了结构与语义一致性，可实现真值标签的自动继承，无需额外人工标注。该数据集具备丰富的光照条件、目标尺度与目标分布多样性，并包含9类挑战因子的标注信息（BCO、SCO、MCO、CB、OV、OC、TC、IC、LI）。这使得EVIAC成为一款面向复杂伪装条件下多模态感知研究的多样化、可靠且成本高效的基准数据集。

搜集汇总

背景与挑战

背景概述

EVIAC是一个大规模可见光-红外伪装目标检测增强基准数据集，包含6000个精确对齐的图像对，覆盖四种场景环境。它通过风格迁移算法增强多样性，自动继承标注并包含9种挑战因素注释，为复杂伪装条件下的多模态感知研究提供多样、可靠且高效的基准。