FCD-78M-2

# FCD-78M-2
FCD-78M is a large-scale, multi-task, instruction-tuning dataset in the field of remote sensing.

This is the second part of FCD-78M. The first part can be found in [FCD-78M](https://www.modelscope.cn/datasets/TianHuiLab/FCD-78M/)

The tasks include in this dataset are: Task06_REG_CLS_HBB, Task07_REG_CLS_OBB, Task09_REG_DET_HBB, Task10_REG_DET_OBB, Task12_PIX_CLS, Task13_PIX_SEG

Due to the excessively large size of some tasks, we have split them. Please merge the split files before decompressing, such as
```
cat Task01_* > Task01_IMG_CLS.zip
```

## Key Features
- 14 tasks
- 5.6 million images
- 78 million annotations

## Download the Dataset

GIT
```
# Make sure you have git-lfs installed (https://git-lfs.com)
git lfs install
git clone https://www.modelscope.cn/datasets/TianHuiLab/FCD-78M.git

```

SDK
```
from modelscope.msdatasets import MsDataset
ds = MsDataset.load('TianHuiLab/FCD-78M')
```

## Dataset Structure
```

FCD-78M/
│
├── README.md # Dataset introduction and usage instructions
└── FCD/
├── json_train_taskall/ # This is a folder of all task training annotations
│ ├── train_task14_all.json # Single-instruction Annotations
│ └── train_task14_all_multi-instructions-version.json # Multi-instruction Annotations
│
├── Task01_IMG_CLS.zip/ # Each zip file contain one of 14 tasks
│ ├── train/ #
│ │ ├── Images/ # Training images
│ │ │ ├── train_IMG_CLS_00000001.png
│ │ │ └── ...
│ │ │
│ │ └── Annotations/ # Training annotations for the task
│ │ ├── train_IMG_CLS.json # Single-instruction Annotations for the task
│ │ └── train_IMG_CLS_multi-instructions-version.json # Multi-instruction Annotations for the task
│ └── test/
│ ├── AID/ # 67 Testing datasets
│ │ ├── Images/ # Testing images
│ │ │ ├── test_IMG_CLS_00000001.png
│ │ │ └── ...
│ │ └── Annotations/ # Testing annotations
│ │ └── test_IMG_CLS.json
│ └──...
└── ...

# FCD-78M-2
FCD-78M是遥感领域的大规模多任务指令微调（Instruction Tuning）数据集。

本数据集为FCD-78M的第二部分，第一部分可于[FCD-78M](https://www.modelscope.cn/datasets/TianHuiLab/FCD-78M/)获取。

本数据集包含的任务如下：Task06_REG_CLS_HBB、Task07_REG_CLS_OBB、Task09_REG_DET_HBB、Task10_REG_DET_OBB、Task12_PIX_CLS、Task13_PIX_SEG。

由于部分任务数据集体量过大，我们对其进行了拆分。请在解压前合并拆分后的文件，例如执行如下命令：

cat Task01_* > Task01_IMG_CLS.zip

## 核心特性
- 共计14项任务
- 560万张图像
- 7800万条标注

## 数据集下载
### Git 方式

# 请确保已安装 git-lfs（https://git-lfs.com）
git lfs install
git clone https://www.modelscope.cn/datasets/TianHuiLab/FCD-78M.git

### SDK 方式

from modelscope.msdatasets import MsDataset
ds = MsDataset.load('TianHuiLab/FCD-78M')

## 数据集结构

FCD-78M/
│
├── README.md # 数据集介绍与使用说明
└── FCD/
├── json_train_taskall/ # 全任务训练标注文件夹
│ ├── train_task14_all.json # 单指令标注文件
│ └── train_task14_all_multi-instructions-version.json # 多指令标注文件
│
├── Task01_IMG_CLS.zip/ # 每个压缩包对应14项任务中的一项
│ ├── train/ # 训练集目录
│ │ ├── Images/ # 训练图像
│ │ │ ├── train_IMG_CLS_00000001.png
│ │ │ └── ...
│ │ │
│ │ └── Annotations/ # 该任务的训练标注文件
│ │ ├── train_IMG_CLS.json # 该任务的单指令标注文件
│ │ └── train_IMG_CLS_multi-instructions-version.json # 该任务的多指令标注文件
│ └── test/
│ ├── AID/ # 67个测试数据集
│ │ ├── Images/ # 测试图像
│ │ │ ├── test_IMG_CLS_00000001.png
│ │ │ └── ...
│ │ └── Annotations/ # 测试标注文件
│ │ └── test_IMG_CLS.json
│ └──...
└── ...