3d_optical_flow_droid

# 3D Optical Flow DROID Dataset

Processed DROID robotics dataset with optical flow and scene flow annotations.

## Dataset Structure

Organized by lab, each trajectory in separate tar.gz archive:

```
IPRL/IPRL+2023-06-19+Mon_Jun_19_23:27:48_2023.tar.gz
CLVR/CLVR+2023-...tar.gz
... (15 labs, ~33K trajectories)
```

Each trajectory contains:
- `metadata.json` - Trajectory metadata
- `trajectory.h5` - Robot state and actions
- `camera_left/`, `camera_right/` - Camera data
- `rgb/` - RGB images
- `depth/` - Depth maps
- `optical_flow_with_mask/` - 2D optical flow
- `scene_flow/` - 3D scene flow

## Usage

```python
from huggingface_hub import hf_hub_download
import tarfile

# Download specific trajectory
tar_path = hf_hub_download(
repo_id="Salesforce/3d_optical_flow_droid",
filename="IPRL/IPRL+2023-06-19+Mon_Jun_19_23:27:48_2023.tar.gz",
repo_type="dataset"
)

# Extract
with tarfile.open(tar_path, "r:gz") as tar:
tar.extractall("./data")
```

## Stats

- Trajectories: ~33,108
- Size: ~26 TB (compressed)
- Labs: 15 robotics labs
- Frames: ~600-700 per trajectory

## Citation

```bibtex
@article{droid2024,
title={DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset},
year={2024}
}
```

# 3D光流DROID数据集（3D Optical Flow DROID Dataset）

经过预处理的DROID机器人数据集，附带光流（optical flow）与场景流（scene flow）标注。

## 数据集组织方式
本数据集按实验室进行分类存储，每条轨迹均封装于独立的tar.gz压缩归档中：

IPRL/IPRL+2023-06-19+Mon_Jun_19_23:27:48_2023.tar.gz
CLVR/CLVR+2023-...tar.gz
……（共15个实验室，约33000条轨迹）

每条轨迹包含以下文件与目录：
- `metadata.json`：轨迹元数据文件
- `trajectory.h5`：存储机器人状态与动作数据的文件
- `camera_left/`、`camera_right/`：左右相机数据目录
- `rgb/`：RGB图像目录
- `depth/`：深度图目录
- `optical_flow_with_mask/`：带掩码的二维光流数据目录
- `scene_flow/`：三维场景流数据目录

## 使用方法
python
from huggingface_hub import hf_hub_download
import tarfile

# 下载指定轨迹
tar_path = hf_hub_download(
repo_id="Salesforce/3d_optical_flow_droid",
filename="IPRL/IPRL+2023-06-19+Mon_Jun_19_23:27:48_2023.tar.gz",
repo_type="dataset"
)

# 解压数据
with tarfile.open(tar_path, "r:gz") as tar:
tar.extractall("./data")

## 数据集统计
- 轨迹总数：约33108条
- 压缩后总容量：约26 TB
- 参与实验室：15个机器人学实验室
- 单轨迹帧数：约600至700帧

## 引用格式
bibtex
@article{droid2024,
title={DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset},
year={2024}
}