JianhuiWei/UniVBench

## Dataset Download

To download the whole UniVBench dataset, run the following command in your terminal:

```bash
python ./download.py
```

## UniVBench Benchmark Directory Structure

Here is the directory structure for `UniVBench/benchmark` along with descriptions for each folder and file type:

```text
UniVBench/
├── full_list/ # Complete benchmark data for video editing tasks
│ ├── 1/ # Each numbered folder represents a specific test case
│ │ ├── 1.mp4 # The original source video file
│ │ ├── caption.json # Captions for the video **[T2V, V2T, V2V tasks]**
│ │ ├── number_of_shot.txt # Records the number of shots in the video
│ │ ├── text_editing.txt # Text editing instructions (original instruction)
│ │ ├── text_editing_en.txt # English text editing **[TV2V]** instructions (Optimized instruction; English)
│ │ ├── text_editing_cn.txt # Chinese text editing **[TV2V]** instructions (Optimized instruction; Chinese)
│ │ ├── reference_editing_single/
│ │ │ ├── reference_editing.txt # Editing instructions for the single reference image (original instruction)
│ │ │ ├── reference_editing_en.txt # English editing instructions with single reference image**[RV2V]** (Optimized instruction)
│ │ │ ├── reference_editing_cn.txt # Chinese editing instructions with single reference image**[RV2V]** (Optimized instruction)
│ │ │ └── reference_image_1.jpeg
│ │ └── reference_editing_multiple/ # Editing materials directory for multiple reference images
│ │ ├── reference_editing.txt # Editing instructions for multiple reference images (original instruction)
│ │ ├── reference_editing_en.txt # English editing instructions with multiple reference images**[RV2V]** (Optimized instruction)
│ │ ├── reference_editing_cn.txt # Chinese editing instructions with multiple reference images**[RV2V]** (Optimized instruction)
│ │ ├── reference_image_1.png
│ │ └── reference_image_2.jpeg
│ ├── 2/
│ │ └── ...
│ └── ... (Contains hundreds of similarly numbered test cases)
│
├── R2V/ # Data for Reference-to-Video generation tasks **[R2V]**
│ ├── 1/ # Each numbered folder represents an R2V test case
│ │ ├── R2V_planning.json # Planning configuration and parameters for the R2V task
│ │ ├── reference_image_0.png # Reference image 0 used for video generation
│ │ └── reference_image_1.jpeg # Reference image 1 used for video generation
│ ├── 2/
│ │ └── ...
│ └── ... (Contains hundreds of R2V test cases)
│
└── reference_images_set/ # A categorized repository of reference images
├── animals/ # Reference images of animals
├── objects/ # Reference images of objects
├── people/ # Reference images of people
├── plants/ # Reference images of plants
└── scenes/ # Reference images of scenes
```

### 数据集下载
若需下载完整的UniVBench数据集，请在终端中执行以下命令：
bash
python ./download.py

### UniVBench 基准测试目录结构
以下为`UniVBench/benchmark`的目录结构，以及各文件夹与文件类型的详细说明：
text
UniVBench/
├── full_list/ # 涵盖视频编辑任务的完整基准测试数据集
│ ├── 1/ # 每个带编号的文件夹代表一个特定测试用例
│ │ ├── 1.mp4 # 原始源视频文件
│ │ ├── caption.json # 视频字幕（适用于**[文本到视频(Text-to-Video, T2V)、视频到文本(Video-to-Text, V2T)、视频到视频(Video-to-Video, V2V)]**任务）
│ │ ├── number_of_shot.txt # 记录视频中的镜头数量
│ │ ├── text_editing.txt # 文本编辑指令（原始指令）
│ │ ├── text_editing_en.txt # 英文文本编辑指令（优化后指令；英文版本）**[TV2V]**
│ │ ├── text_editing_cn.txt # 中文文本编辑指令（优化后指令；中文版本）**[TV2V]**
│ │ ├── reference_editing_single/
│ │ │ ├── reference_editing.txt # 单参考图像的编辑指令（原始指令）
│ │ │ ├── reference_editing_en.txt # 带单参考图像的英文编辑指令（优化后指令）**[参考视频到视频(Reference-based Video-to-Video, RV2V)]**
│ │ │ ├── reference_editing_cn.txt # 带单参考图像的中文编辑指令（优化后指令）**[RV2V]**
│ │ │ └── reference_image_1.jpeg
│ │ └── reference_editing_multiple/ # 多参考图像编辑材料目录
│ │ ├── reference_editing.txt # 多参考图像的编辑指令（原始指令）
│ │ ├── reference_editing_en.txt # 带多参考图像的英文编辑指令（优化后指令）**[RV2V]**
│ │ ├── reference_editing_cn.txt # 带多参考图像的中文编辑指令（优化后指令）**[RV2V]**
│ │ ├── reference_image_1.png
│ │ └── reference_image_2.jpeg
│ ├── 2/
│ │ └── ...
│ └── ... (包含数百个同类编号测试用例)
│
├── R2V/ # 参考图像到视频生成任务数据集**[Reference-to-Video generation, R2V]**
│ ├── 1/ # 每个带编号的文件夹代表一个R2V测试用例
│ │ ├── R2V_planning.json # R2V任务的规划配置与参数
│ │ ├── reference_image_0.png # 用于视频生成的参考图像0
│ │ └── reference_image_1.jpeg # 用于视频生成的参考图像1
│ ├── 2/
│ │ └── ...
│ └── ... (包含数百个R2V测试用例)
│
└── reference_images_set/ # 分类整理的参考图像资源库
├── animals/ # 动物类参考图像
├── objects/ # 物体类参考图像
├── people/ # 人物类参考图像
├── plants/ # 植物类参考图像
└── scenes/ # 场景类参考图像