# Dataset licences

AnyLearning OSS is distributed under Apache-2.0. Dataset files remain under
their own terms, so every dataset in this repository needs a licence that
permits commercial use and redistribution.

Record the licence **and its source** for each dataset. "It was on the internet"
is not a licence.

## Cleared for use

| Dataset | Task | Licence | Source |
|---|---|---|---|
| `zhanglabdata_chest_xray` | Image classification | **CC BY 4.0** | Kermany, Zhang & Goldbaum (2018), [Cell](https://www.cell.com/cell/fulltext/S0092-8674(18)30154-5) |
| `helmet_jacket` | Object detection | **Apache 2.0** | [Dataset Ninja](https://datasetninja.com/safety-helmet-and-reflective-jacket#license) |
| `electron_microscopy_particle_segmentation` | Semantic segmentation | **CC BY 4.0** | [Dataset Ninja](https://datasetninja.com) |
| `american_sign_language` | Handpose classification | **Public Domain** | [Roboflow Public](https://public.roboflow.com/object-detection/american-sign-language-letters/1) |
| `keypoint_stick_figures` | Keypoint detection | **CC0 1.0** | Generated by the AnyLearning project; no third-party pixels |
| `desert_locust_keypoints` | Keypoint detection | **Apache 2.0** | Graving et al. (2019), [official DeepPoseKit-Data repository](https://github.com/jgraving/DeepPoseKit-Data) |
| `vertebral_keypoints` CDN subset | Keypoint detection | **CC BY 4.0** | Reyes (2025), [Mendeley Data V1](https://doi.org/10.17632/5jdfdgp762.1) |
| `structured/v1/bank-marketing.csv` | Tabular classification | **CC BY 4.0** | Moro, Rita & Cortez (2014), [UCI DOI 10.24432/C5K306](https://doi.org/10.24432/C5K306) |
| `structured/v1/concrete-strength.csv` | Tabular regression | **CC BY 4.0** | Yeh (1998), [UCI DOI 10.24432/C5PK67](https://doi.org/10.24432/C5PK67) |
| `structured/v1/banking77.csv` | Text classification | **CC BY 4.0** | Casanueva et al. (2020), [PolyAI task-specific datasets](https://github.com/PolyAI-LDN/task-specific-datasets) |
| `structured/v1/truthfulqa-evaluation.csv` | Response evaluation | **Apache 2.0** | Lin, Hilton & Evans (2022), [official TruthfulQA repository](https://github.com/sylinrl/TruthfulQA) |

CC BY 4.0 requires **attribution**, which is satisfied by each dataset's README
and this table. It does not restrict commercial use and has no ShareAlike term.

## Fetch-on-demand datasets

`vertebral_keypoints` contains instructions but no source pixels in Git. The
Spondylolisthesis Vertebral Landmark v1 record publishes the archive under
CC BY 4.0 while describing a mixture of locally collected and BUU-LSPINE source
images. A versioned 64/16-image engineering subset is therefore mirrored on the
CDN with `ATTRIBUTION.md` embedded in both archives. The complete dataset stays
fetch-on-demand from its DOI, and neither the subset nor its measured result is
clinical validation.

## Excluded datasets awaiting a licence decision

The following collections existed in the source catalog but are not included in
this public snapshot because no redistribution licence was recorded. They must
not be added until their upstream terms are confirmed.

| Dataset | Task | Status |
|---|---|---|
| `neu_surface_defect` | Image classification | **Excluded.** No licence was recorded. NEU-DET originates from Northeastern University and is commonly described for academic/research use, which is not a redistribution grant. |
| `dental_segment` | Instance segmentation | **Excluded.** Its source export had no licence line. Confirm the licence on the upstream project before adding it. |

## Adding a dataset

1. Confirm the licence permits commercial use **and** redistribution.
2. Reject anything non-commercial (CC BY-NC, "research only", "academic use").
3. Treat ShareAlike (CC BY-SA) with care: it is usable, but derivatives inherit
   the term. Prefer fetch-on-demand over committing such data here.
4. Add a `README.md` to the dataset folder recording licence, source URL, task,
   and any citation the licence requires.
5. Add a row to the table above.