"RS-AirSim"

"To bridge the color discrepancy between adversarial textures and surrounding environments during digital training, we constructed an environment color extraction and fusion module. For this purpose, we first generated 1,260 base training images. These images were then combined with nine different texture colors, resulting in 11,340 realistic reference images. Furthermore, to support adversarial texture training, we generated an additional 10,080 adversarial images, including 7,560 for training and 2,520 for testing. These images span nine weather conditions, 24 viewing angles, and seven flight altitudes. All images are of 800 \u00d7 800 resolution and are accompanied by detailed object annotations and corresponding mask images, ensuring both diversity and applicability. We refer to this dataset as RS-AirSim."