SoccerSynth-Field
收藏资源简介:
SoccerSynth-Field数据集是由名古屋大学信息学研究科和理化学研究所先进智能研究中心的研究人员开发的合成足球场地数据集。该数据集通过Unreal Engine 5模拟器生成,旨在提供一种灵活高效的数据收集替代方案。数据集包含20,000张图像,通过模拟不同的光照、草坪纹理、摄像机角度等条件,增加了数据集的多样性。在数据集中,通过在图像中添加假的场地线条来引入视觉噪声,并模拟真实场景中的条件,测试模型的鲁棒性。该数据集的创建是为了提高场地检测模型的准确性和鲁棒性,特别是在真实世界数据标注有限的环境中。
SoccerSynth-Field is a synthetic soccer field dataset developed by researchers from the Graduate School of Informatics at Nagoya University and the RIKEN Center for Advanced Intelligence Project. Generated via the Unreal Engine 5 simulator, this dataset aims to provide a flexible and efficient alternative for data collection. It contains 20,000 images, with diversity enhanced by simulating diverse conditions including varying lighting, lawn textures, camera angles and more. Visual noise is introduced by adding fake field lines to the images, and real-world scene conditions are simulated to test model robustness. This dataset was constructed to improve the accuracy and robustness of field detection models, particularly in environments where real-world data annotations are limited.




