anime_head_detection

Dataset for anime head detection (include the entire head, not only the face parts).

| Dataset | Train | Test | Validate | Description |
|------------------------|-------|------|----------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| v2.0 | 16050 | 766 | 1528 | High-quality manually annotated head detection dataset, containing images from bangumi and a large number of complex multi-person illustrations, which can significantly improve the training quality of the v1.0 dataset. Recommended to be used together with v1.0 dataset. Thanks to [@SumomoLee](https://huggingface.co/SumomoLee) and [@Crystal427](https://huggingface.co/Crystal427) for participating in the annotation task. |
| v2.0-raw | 5350 | 766 | 1528 | Original archive of v2.0, without any preprocessing or data augmentation. |
| ani_face_detection.v1i | 25698 | 113 | 253 | A high-quality third-party dataset (seems to no longer be publicly available, please contact me for removal if it infringes your rights) that can be used for training directly. Although its name includes `face`, but what it actually annotated are `head`. |
| v1.0 | 20889 | 1036 | 2000 | High-quality manually annotated head detection data. Includes images sampled from gelbooru and fancaps. All heads with human characteristics (including front, side, and back views) are marked, while non-human heads without significant humanoid features are not considered as "heads". |
| v1.0-raw | 6963 | 1036 | 2000 | Original archive of v1.0, without any preprocessing or data augmentation. |

We provide an [online demo](https://huggingface.co/spaces/deepghs/anime_object_detection) here.

本数据集面向动漫头部检测（head detection）任务，需标注完整头部，而非仅面部区域。

| 数据集名称 | 训练集 | 测试集 | 验证集 | 数据集描述 |
| --- | --- | --- | --- | --- |
| v2.0 | 16050 | 766 | 1528 | 高质量人工标注头部检测数据集，数据来源于番剧（bangumi）及大量复杂多人插画，可显著提升v1.0版本数据集的训练效果，推荐与v1.0版本数据集配合使用。感谢[@SumomoLee](https://huggingface.co/SumomoLee)与[@Crystal427](https://huggingface.co/Crystal427)参与本次标注工作。 |
| v2.0-raw | 5350 | 766 | 1528 | v2.0版本的原始归档文件，未经过任何预处理或数据增强操作。 |
| ani_face_detection.v1i | 25698 | 113 | 253 | 高质量第三方数据集（目前似乎已不再公开提供，若侵犯您的权益请联系我移除），可直接用于模型训练。尽管其名称包含"face"字样，但实际标注的目标为头部（head）。 |
| v1.0 | 20889 | 1036 | 2000 | 高质量人工标注头部检测数据集，数据采样自gelbooru与fancaps平台。所有具备人类特征的头部（包括正面、侧面与背面视角）均会被标注，而不具备类人特征的非人类头部则不被视为‘头部’。 |
| v1.0-raw | 6963 | 1036 | 2000 | v1.0版本的原始归档文件，未经过任何预处理或数据增强操作。 |

我们在此提供了[在线演示](https://huggingface.co/spaces/deepghs/anime_object_detection)。