README.md

---
license: odbl
task_categories:
- image-classification
tags:
- art
- not-for-all-audiences
- anime
- visual-novel
- nsfw
- vndb
size_categories:
- 100K<n<1M
---
# Dataset Card for VNDB IMG

## Dataset Description

This is a 🤗 Datasets loader for the [vndb.org](https://vndb.org) image database dump.

It contains anime-style images flagged by users according to these categories:

* sexual content: safe/suggestive/explicit
* violence: tame/violent/brutal

## Loading Instructions

For licensing and "moral" reasons, the database dump has to be downloaded manually.

Please download the vndb.org database dump manually from <https://vndb.org/d14>.
Download the "Near-complete database" `vndb-db-latest.tar.zst` file.
Use `rsync` to download the 'Images' collection.

Create the following directory structure:

```
my/dataset/path
├── db
│   └── vndb-db-latest.tar.zst
└── vndb-img # this is the directory you downloaded with rsync
    ├── ch
    ├── cv
    ├── sf
    ├── st
    └── ...
```

Inside `my/dataset/path/db` run

```
zstd -d vndb-db-latest.tar.zst
```

and

```
tar -xf vndb-db-latest.tar
```

The final directory structure should look like this:

```
my/dataset/path
├── db
│   ├── vndb-db-latest.tar
│   ├── vndb-db-latest.tar.zst
│   ├── db
│   └── ...
└── vndb-img
    ├── ch
    ├── cv
    ├── sf
    ├── st
    └── ...
```

Finally, load the dataset

```python
datasets.load_dataset('carbon225/vndb_img', data_dir='my/dataset/path')
```

## Dataset Structure

The following fields are provided:
```python
{
    'index': datasets.Value('int32'),
    'id': datasets.Value('string'),
    'width': datasets.Value('int32'),
    'height': datasets.Value('int32'),
    'c_votecount': datasets.Value('int32'),
    'c_sexual_avg': datasets.Value('int32'),
    'c_sexual_stddev': datasets.Value('int32'),
    'c_violence_avg': datasets.Value('int32'),
    'c_violence_stddev': datasets.Value('int32'),
    'c_weight': datasets.Value('int32'),
    'type': datasets.ClassLabel(names=['character', 'cover', 'screenshot_full', 'screenshot_thumb']),
    'sexual_class': datasets.ClassLabel(names=['safe', 'suggestive', 'explicit']),
    'violence_class': datasets.ClassLabel(names=['tame', 'violent', 'brutal']),
    'file_name': datasets.Value('string'),
    'full_path': datasets.Value('string'),
    'image': datasets.Image(),
}
```

## Supported Tasks

With a few modifications the data can be used for:
* image classification of NSFW material
* image generation/super-resolution/...
* ...

## Considerations for Using the Data

The images are ***hardcore***, to say the least. I recommend not looking.

## Licensing Information

Using this dataset requires the user to download data manually from vndb.org.

All information on VNDB is made available under the Open Database License.
Any rights in individual contents of the database are licensed under the Database Contents License.

With the following exceptions:
* Anime data is obtained from the AniDB.net UDP API and is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0.
* Images, visual novel descriptions and character descriptions are gathered from various online sources and may be subject to separate license conditions.