wanghaofan/pokemon-wiki-captions

---
dataset_info:
features:
- name: image
dtype: image
- name: name_en
dtype: string
- name: name_zh
dtype: string
- name: text_en
dtype: string
- name: text_zh
dtype: string
splits:
- name: train
num_bytes: 117645424.0
num_examples: 898
download_size: 117512478
dataset_size: 117645424.0
---
# Dataset Card for Pokémon wiki captions

This project is inspired by [pokmon-blip-captions](https://huggingface.co/datasets/lambdalabs/pokemon-blip-captions), where the captions are all generated by pre-trained BLIP without any manual effort.
However, the quality and accuracy of their captions are not satisfactory enough, which leaves it known whether better captions lead to better results. This motivates our dataset.

# Example

> General attribute, looks like a little monkey, body color is composed of purple and beige, the end of the tail is like a hand

> Poisonous attributes, it looks like a huge purple cobra, with black stripes on its body, small head, and triangular eyes

# Properties

All 898 images are from [The Complete Pokemon Images Data Set](https://www.kaggle.com/datasets/arenagrenade/the-complete-pokemon-images-data-set?resource=download) in Kaggle with size 475x475. Each image is accompanied with corresponding
pokemon name and its detailed description from [Pokemon Wiki](https://wiki.52poke.com/wiki/%E4%B8%BB%E9%A1%B5), English and Chinese captions are provided. Human efforts are also involved to revise.

# How to use
```
from datasets import load_dataset

dataset = load_dataset("wanghaofan/pokemon-wiki-captions")
```

The dataset is formatted as below. For each row the dataset contains `image`, `name_en`, `name_zh`, `text_en` and `text_zh` keys. `image` is a varying size PIL jpeg, `name` is the name of pokemon, and `text` is the accompanying text caption. Only a train split is provided.

```
DatasetDict({
train: Dataset({
features: ['image', 'name_en', 'name_zh', 'text_en', 'text_zh'],
num_rows: 898
})
})

```

# Citation
If you use this dataset in your work, please cite it as:

```
@misc{wanghaofan2022pokemon,
author = {Haofan Wang},
title = {Pokemon wiki captions},
year={2022},
howpublished= {\url{https://huggingface.co/datasets/wanghaofan/pokemon-wiki-captions/}}
}
```

原始信息汇总

数据集概述

数据集信息

• 特征:
  • image: 图像数据
  • name_en: 英文名称，字符串类型
  • name_zh: 中文名称，字符串类型
  • text_en: 英文描述，字符串类型
  • text_zh: 中文描述，字符串类型
• 分割:
  • train: 训练集，包含 898 个样本，总大小为 117645424.0 字节
• 下载大小: 117512478 字节
• 数据集大小: 117645424.0 字节

数据集描述

• 图像来源: 所有 898 张图像来自 Kaggle 上的 The Complete Pokemon Images Data Set，每张图像尺寸为 475x475。
• 描述信息: 每张图像附有对应的宝可梦名称和详细描述，描述来自 Pokemon Wiki，提供英文和中文两种语言的描述，并有人工修订。

使用方法

python from datasets import load_dataset

dataset = load_dataset("wanghaofan/pokemon-wiki-captions")

数据集格式

• 数据集结构: python DatasetDict({ train: Dataset({ features: [image, name_en, name_zh, text_en, text_zh], num_rows: 898 }) })

引用

如果使用此数据集，请按以下格式引用：

@misc{wanghaofan2022pokemon, author = {Haofan Wang}, title = {Pokemon wiki captions}, year={2022}, howpublished= {url{https://huggingface.co/datasets/wanghaofan/pokemon-wiki-captions/}} }