Xiao215/pixiv-image-with-caption

---
title: "Pixiv Daily Trending Illusions Dataset"
language:
- "en"
license:
- "unknown" # Since the exact terms for redistributed scraped content are unclear
multilinguality:
- "monolingual"
size_categories:
- "100<n<1K" # This dataset contains 949 images
source_datasets:
- "extended|other-Pixiv" # Indicating that this dataset extends from or is based on Pixiv
task_categories:
- "image-to-text"
- "text-to-image"
task_ids:
- "image-captioning"
annotations_creators:
- "machine-generated"
---

# Dataset Card for Pixiv Daily Trending Illusions Dataset

Note, this dataset contains copyright issue, and is displayed for fun personal project only. Do not use it.

## Table of Contents
- [Dataset Description](#dataset-description)
- [Access](#access)
- [Dataset Structure](#dataset-structure)
- [Usage](#usage)
- [Acknowledgements](#acknowledgements)
- [Licensing](#licensing)

## Dataset Description

- **Homepage:** [Pixiv Daily Trending Illusions](https://www.pixiv.net/discovery?mode=safe)
- **Repository:** [HuggingFace Dataset](https://huggingface.co/datasets/Xiao215/pixiv-image-with-caption)
- **Paper:** N/A
- **Leaderboard:** N/A

### Dataset Summary

This dataset comprises 949 images scrapped from Pixiv's daily trend, specifically curated to include only illustrations that are illusions and suitable for all ages. Each image in the dataset is accompanied by a caption generated by the LLaVa model, providing a descriptive or interpretive text element for the visual content.

### Languages

Captions are generated in [Language(s)], as processed by the LLaVa model on HuggingFace.

## Access

The dataset can be accessed through the HuggingFace `datasets` library using the following code snippet:

```python
from datasets import load_dataset

dataset = load_dataset("Xiao215/pixiv-image-with-caption")
```
## Dataset Structure

### Data Instances

A data instance in this dataset comprises the following fields:
- `image_name`: a `string` representing the filename of the image, following the pattern `pixiv{image_id}.png`.
- `caption`: a `string` generated by the LLaVa model, describing or interpreting the image.

Example:
```python
{
"image_name": "pixiv100028371.png",
"caption": "A mesmerizing pattern that appears to swirl endlessly."
}
```

### Data Splits

This dataset is provided in a single split:
- The `all` split contains all 949 images and their corresponding captions.

## Usage

This dataset can be used for tasks such as image captioning, visual understanding, and training models to generate descriptive texts for abstract visual content. Here's an example of how to load and use the dataset:

from datasets import load_dataset

# Example usage
```python
from datasets import load_dataset
dataset = load_dataset("Xiao215/pixiv-image-with-caption")

for sample in dataset['all']:
print(sample['image_name'], sample['caption'])
```
## Usage with cache
```python
from datasets import load_dataset
# Specify the path where you want to cache the dataset
cache_dir = "/path/to/your/desired/cache/directory"

# Load the dataset and specify the cache directory
dataset = load_dataset("Xiao215/pixiv-image-with-caption", cache_dir=cache_dir)
```
## Acknowledgements

This dataset was collected from [Pixiv](https://www.pixiv.net/discovery?mode=safe), with captions generated by the [LLaVa model](https://huggingface.co/docs/transformers/en/model_doc/llava) on HuggingFace.

## Licensing

Please review Pixiv's terms of use and licensing information to ensure compliance when using this dataset. The use of the LLaVa model for generating captions is subject to the terms and conditions provided by HuggingFace and the model's authors.

原始信息汇总

Pixiv Daily Trending Illusions Dataset 概述

数据集基本信息

• 标题: Pixiv Daily Trending Illusions Dataset
• 语言: 单一语言（具体语言未指定）
• 许可证: 未知
• 多语言性: 单语
• 大小: 949 张图像
• 来源数据集: 扩展自 Pixiv
• 任务类别:
  • 图像到文本
  • 文本到图像
• 任务ID: 图像标题生成
• 标注创建者: 机器生成

数据集描述

• 数据集概要: 包含949张从Pixiv每日趋势中筛选的适合所有年龄的幻觉图像。每张图像均配有由LLaVa模型生成的描述性或解释性文本。
• 语言: 标注文本由LLaVa模型处理，具体语言未指定。

数据集结构

• 数据实例: 每个实例包含以下字段：
  • image_name: 图像文件名，格式为pixiv{image_id}.png。
  • caption: 由LLaVa模型生成的图像描述文本。
• 数据分割: 单一分割，包含所有949张图像及其对应标题。

使用方法

• 示例代码: 通过HuggingFace datasets 库加载数据集，并可用于图像标题生成、视觉理解和训练模型生成描述文本等任务。

许可证

• 使用此数据集需遵守Pixiv的条款和许可证，以及HuggingFace和LLaVa模型作者的条款和条件。