---
license: cc-by-nc-4.0
task_categories:
  - text-to-video
  - image-to-video
language:
  - zh
tags:
  - automotive
  - car
  - video-dataset
  - diffusion
  - rigid-body
  - i2v
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: "Circle_Rotate/*.mp4"
---

# Circle-Rotate Dataset

<div align="center">

**A High-Quality Video Dataset for Rigid-Body Video Generation with Implicit Motion Alignment**

[![License: CC BY-NC 4.0](https://img.shields.io/badge/License-CC%20BY--NC%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc/4.0/)
[![Dataset Size](https://img.shields.io/badge/Size-~10GB-blue.svg)]()
[![Videos](https://img.shields.io/badge/Videos-2168-green.svg)]()

</div>

---

## Overview

This dataset contains **2,168 professionally produced videos** with implicit geometric alignment, designed for training video generation models. Each video features smooth circular camera motion around a static subject, providing consistent rigid-body rotation patterns.

**Key Applications:**

- **Video Diffusion Models** (e.g., Wan2.1, Wan2.2, CogVideoX, HunyuanVideo)
- **Image-to-Video (I2V) Generation**
- **Camera Motion Control Learning**
- **Rigid-Body Video Generation**

---

## Dataset Structure

```
circle-rotate-dataset/
├── Circle_Rotate/          # Main video dataset (~10GB)
│   ├── *.mp4              # 2168 video files
│   └── *.txt              # Corresponding caption files
│
└── Gen/                    # First/Last frame pairs for I2V
    ├── Static/            # Static scene videos (29 pairs)
    │   ├── first_frame/
    │   └── last_frame/
    └── Dynamic/           # Dynamic scene videos (30 pairs)
        ├── first_frame/
        └── last_frame/
```

---

## Related Project

This dataset is part of the research project **"Implicit Motion Alignment: A Data-Centric Empirical Study for Rigid-Body Video Generation"**.

- **GitHub**: [Circle-Rotate](https://github.com/Jklaity/Circle-Rotate)
- **Model**: [circle-rotate-lora](https://huggingface.co/jk1741391802/circle-rotate-lora)

---

## Data Format

### Video Naming Convention

Each video filename contains rich metadata separated by `&`:

```
{时间}&{展示类型}&{场景}&{视角}&{景别}&{默认}&{环绕方向}&{位置}&{型号}&{车衣}&{特殊样式}.mp4
```

| Field | Description | Examples |
|-------|-------------|----------|
| 时间 | Time of day | 夜晚 (Night), 白天 (Day) |
| 展示类型 | Display type | 静态展示 (Static) |
| 场景 | Scene/Environment | 都市夜景, 展厅, 户外 |
| 视角 | Camera angle | 侧面, 前脸, 车顶, 车身线条 |
| 景别 | Shot type | 中景, 全景, 远景, 大景 |
| 环绕方向 | Camera direction | 环绕右左, 环绕左右 |
| 位置 | Position | 右侧, 左侧 |

### Caption Format

Each `.txt` file contains a Chinese description of the video content:

```
汽车侧面静态展示，镜头环绕左右, 中景,场景是都市夜景,夜晚
```

---

## Statistics

| Category | Count |
|----------|-------|
| Total Videos | **2,168** |
| Total Size | **~10 GB** |
| Video Format | MP4 |
| Caption Format | TXT (Chinese) |
| Gen Static Pairs | 29 |
| Gen Dynamic Pairs | 30 |

---

## Usage

### Load with Hugging Face

```python
from huggingface_hub import snapshot_download

# Download entire dataset
snapshot_download(
    repo_id="jk1741391802/circle-rotate-dataset",
    repo_type="dataset",
    local_dir="./circle-rotate-dataset"
)
```

### Load Specific Folder

```python
from huggingface_hub import hf_hub_download
import os

# Download only Circle_Rotate videos
snapshot_download(
    repo_id="jk1741391802/circle-rotate-dataset",
    repo_type="dataset",
    local_dir="./data",
    allow_patterns="Circle_Rotate/*"
)
```

---

## Applications

### 1. Video Generation Training
Train video diffusion models to generate videos with stable subjects and smooth camera motion, reducing subject drift.

### 2. Image-to-Video (I2V)
Use the `Gen/` folder's first_frame and last_frame pairs for I2V model training with geometric alignment.

### 3. Camera Motion Learning
Learn circular camera trajectories for controllable rigid-body video generation.

---

## License

This dataset is released under the [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) license.

- **Attribution** — You must give appropriate credit
- **NonCommercial** — You may not use the material for commercial purposes

---

## Citation

```bibtex
@dataset{circle_rotate_dataset_2024,
  title={Circle-Rotate Dataset: A Video Dataset for Rigid-Body Video Generation},
  author={jk1741391802},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/jk1741391802/circle-rotate-dataset}
}
```

---

<div align="center">

**Made with ❤️ for the AI Video Generation Community**

</div>