AOT Characters Dataset

The AOT Characters Dataset consists of 4,041 manually collected images of 14 characters from Attack on Titan, captured across diverse scenes and expressions. The data was gathered via HD screenshots and carefully cleaned by removing blurry or obstructed frames. When tested with various deep learning models, results showed competitive accuracy with Deep Danbooru on the full dataset. This demonstrates the dataset’s value for anime-specific tasks like character classification and supports reproducible model evaluation in stylized visual media.AOT角色数据集（AOT Characters Dataset）共包含4041张手动采集的图像，涵盖《进击的巨人》（Attack on Titan）中的14个角色，采集场景与角色表情均丰富多样。该数据集通过高清截图收集得到，并经过严格的清洗处理，剔除了模糊或存在遮挡的画面帧。采用多种深度学习模型对其进行测试后，其在全量数据集上的分类准确率可与Deep Danbooru相媲美。这一结果证实了该数据集在动漫专属任务（如角色分类）中的应用价值，同时可为风格化视觉媒体领域的可复现模型评估提供有力支撑。