|
--- |
|
license: mit |
|
task_categories: |
|
- object-detection |
|
tags: |
|
- art |
|
size_categories: |
|
- 1K<n<10K |
|
--- |
|
|
|
Dataset for anime person detection. |
|
|
|
| Dataset | Train | Test | Validate | Description | |
|
|-------------|-------|------|----------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| |
|
| v1.1 | 9255 | 460 | 877 | Annotated on the Roboflow platform, including labeled data for various types of anime images (e.g. illustrations, comics). The dataset has also undergone data augmentation techniques to enhance its diversity and quality. | |
|
| raw | 3085 | 460 | 877 | The same as `v1.1` dataset, without any preprocess and data augmentation. Suitable for directly upload to Roboflow platform. | |
|
| AniDet3.v3i | 16124 | 944 | 1709 | Third-party dataset, source: https://universe.roboflow.com/university-of-michigan-ann-arbor/anidet3-ai42v/dataset/3 . The dataset only contains images from anime series. This means the models directly trained on it will not perform well on illustrations and comics. | |
|
|
|
The best practice is to combine the `AniDet3.v3i` dataset with the `v1.1` dataset for training. We provide an [online demo](https://huggingface.co/spaces/deepghs/anime_object_detection). |
