megalith-10m-florence2

# Megalith-10M with Florence-2 Caption

[日本語はこちら](README_JA.md)

This reposity is the supplymentary of [Megalith-10M](https://huggingface.co/datasets/madebyollin/megalith-10m).
Megalith-10M is an CC-0 like image dataset. However, the dataset does not contain the image caption.
Therefore, we caption the images by [Florence 2](https://huggingface.co/microsoft/Florence-2-large).

# Usage

```python
from datasets import load_dataset

dataset = load_dataset("aipicasso/megalith-10m-florence2")
```

# How to get images

```bash
git lfs install
git clone https://huggingface.co/datasets/drawthingsai/megalith-10m
```

# Intended Use
- Research Vision & Language
- Develop text-to-image model or image-to-text model.

# Procedure
We captioned the images by Florence-2-large.
We gave the model MORE_DETAILED_CAPTION task.

# Captioning Cost
- 3000 GPUhours (T4)

# Acknowledgement
We appreciate [madebyollin](https://huggingface.co/madebyollin) to create the great dataset.
This dataset leads futher researchs and development.

# 基于Florence-2生成标注的Megalith-10M数据集
[日语版本请参见README_JA.md]

本仓库为[Megalith-10M](https://huggingface.co/datasets/madebyollin/megalith-10m)的补充项目。
Megalith-10M是类CC-0的图像数据集，但该数据集未附带图像标注。因此，我们通过[Florence 2](https://huggingface.co/microsoft/Florence-2-large)为数据集内的图像生成了标注。

# 使用方法
python
from datasets import load_dataset

dataset = load_dataset("aipicasso/megalith-10m-florence2")

# 获取图像方式
bash
git lfs install
git clone https://huggingface.co/datasets/drawthingsai/megalith-10m

# 预期用途
- 开展视觉与语言交叉领域研究
- 开发文本到图像或图像到文本模型

# 标注流程
我们通过Florence-2-large为图像生成标注，并为模型指定了MORE_DETAILED_CAPTION（详细图像描述生成）任务。

# 标注成本
- 总计3000 GPU小时（基于T4显卡）

# 致谢
我们感谢[madebyollin](https://huggingface.co/madebyollin)构建了这一优质数据集，其为后续相关研究与开发提供了重要支撑。