cua-verse/GUI-360

# cua-verse/GUI-360

cua-lite preprocessed version of GUI-360, a desktop GUI dataset (Windows Office workflows at 1036×728). Covers three task types: grounding:action (UI element action prediction), grounding:point (coordinate localization), and understanding (screen parsing).

## Origin

- [https://huggingface.co/datasets/vyokky/GUI-360](https://huggingface.co/datasets/vyokky/GUI-360)

## Schema

Each row has these columns:

| column | type | notes |
|---|---|---|
| `image_ids` | list[string] | content-addressed ids (`<sha256>.<ext>`), used for cross-parquet dedup |
| `images` | list[Image] | bytes embedded at HF push time; matches `image_ids` index-for-index |
| `messages` | list[struct] | OpenAI-style turns with `role` + structured `content` |
| `metadata` | struct | `{platform, task_type, split, others{...}}` |

Coordinate values in `messages` are normalized to `[0, 1000]` integers.

## Layout

```
<platform>/<task_type>/<split>/<variant>.parquet
```

- `platform` ∈ {desktop, mobile, web}
- `task_type` ∈ {understanding, grounding:action, grounding:point, grounding:bbox, trajectory}
- `split` ∈ {train, eval} — `eval` is an in-distribution held-out slice (not used in training)

## Stats

| platform | task_type | variant | train | eval |
|---|---|---|---:|---:|
| desktop | grounding:action | action | 96,394 | 1,972 |
| desktop | grounding:point | point | 77,901 | 1,585 |
| desktop | understanding | screen_parsing | 95,382 | 1,969 |

## Image store

Images are content-addressed by SHA-256 and deduplicated within this repo.
The mapping `image_id → bytes` is embedded in every parquet via
[`datasets.Image`](https://huggingface.co/docs/datasets/image_load). A
reverse script (see [`cua-lite/cua-lite`](https://github.com/cua-lite/cua-lite))
reconstructs a local `image_store/` + reference-only parquets for
re-dedup across datasets.

- Total images: **165,715**
- Store size: **27.62 GB**

## Notes

Coordinates normalized to [0, 1000]. Original 1036×728 screenshots retained.

## License & Citation

See original dataset (vyokky/GUI-360)

See https://huggingface.co/datasets/vyokky/GUI-360