anime_face_detection

Dataset for anime face detection (face only, not the entire head). | Dataset | Train | Test | Validate | Description | |:-----------------------:|:-----:|:----:|:--------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | v1.4 | 12798 | 622 | 1217 | Additional images from different categories have been annotated based on the `v1` dataset. Furthermore, all automatically annotated data samples from the `v1` dataset have been manually corrected. | | v1.4-raw | 4266 | 622 | 1217 | Same as `v1.4`, without any preprocess and data augmentation. Suitable for directly upload to Roboflow platform. | | v1 | 5943 | 293 | 566 | Primarily consists of illustrations, auto-annotated with [hysts/anime-face-detector](https://github.com/hysts/anime-face-detector), and necessary manual corrections is performed. | | raw | 1981 | 293 | 566 | Same as `v1`, without any preprocess and data augmentation. Suitable for directly upload to Roboflow platform. | | Anime Face CreateML.v1i | 4263 | 609 | 1210 | Third-party dataset, source: https://universe.roboflow.com/my-workspace-mph8o/anime-face-createml/dataset/1 | The best practice is to combine the `Anime Face CreateML.v1i` dataset with the `v1.4` dataset for training. We provide an [online demo](https://huggingface.co/spaces/deepghs/anime_object_detection).本数据集用于动漫人脸检测（仅标注人脸区域，不包含完整头部）。 | 数据集版本 | 训练集 | 测试集 | 验证集 | 详细说明 | |:-----------------------:|:-----:|:----:|:--------:|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | v1.4 | 12798 | 622 | 1217 | 本数据集基于v1版本数据集新增了多类别标注图像。此外，已对v1数据集中所有自动标注的数据样本完成人工修正。 | | v1.4-raw | 4266 | 622 | 1217 | 与v1.4版本完全一致，未进行任何预处理与数据增强操作，可直接上传至Roboflow平台。 | | v1 | 5943 | 293 | 566 | 该数据集主要由插画作品构成，通过[hysts/anime-face-detector](https://github.com/hysts/anime-face-detector)完成自动标注，并进行了必要的人工修正。 | | raw | 1981 | 293 | 566 | 与v1版本完全一致，未进行任何预处理与数据增强操作，可直接上传至Roboflow平台。 | | Anime Face CreateML.v1i | 4263 | 609 | 1210 | 第三方数据集，来源：https://universe.roboflow.com/my-workspace-mph8o/anime-face-createml/dataset/1 | 最佳训练实践为将`Anime Face CreateML.v1i`数据集与`v1.4`数据集结合用于模型训练。我们提供了[在线演示](https://huggingface.co/spaces/deepghs/anime_object_detection)。

搜集汇总

背景与挑战

背景概述

该数据集专用于动漫人脸检测，聚焦于面部而非整个头部，包含多个版本（如v1.4和Anime Face CreateML.v1i），其中v1.4在v1基础上扩展了图像类别并进行了手动修正。最佳实践建议结合Anime Face CreateML.v1i与v1.4进行训练，数据集总大小为5.67GB，采用Apache License 2.0许可证。