# Datasets

Per-scene GGPS capture data. Each scene is uploaded as a **single archive** (e.g. `FTP.zip`),
which expands to the openMVG → COLMAP layout consumed by the
[GGPS code repository](https://github.com/Insta360-Research-Team/GGPS).

## Scenes

| Scene | Archive | Panoramas | Notes |
|-------|---------|-----------|-------|
| FTP   | `FTP.zip` | 354  | ERP panoramas + openMVG reconstruction |
| NSC   | `NSC.zip` | 1862 | ERP panoramas + openMVG reconstruction |
| NSK   | `NSK.zip` | 576  | ERP panoramas + openMVG reconstruction |

## Archive contents

```text
<scene>/
├── images/                  # ERP panoramas (.png)
└── reconstruction/
    ├── sfm_data.bin         # openMVG binary
    ├── sfm_data.json        # openMVG JSON (re-exportable from .bin)
    ├── colorized.ply        # colorized sparse point cloud (3DGS init)
    ├── cloud_and_poses.ply
    └── *_Report.html / *.svg
```

## Usage

Download and extract a scene, then point a config's `source_path` at it and run the
data-preparation pipeline (`scripts_new/prepare_data.sh`) from the code repo.

```bash
hf download Insta360-Research/GGPS datasets/FTP.zip --repo-type model --local-dir .
unzip datasets/FTP.zip -d data/
```