---
pretty_name: AnyLearning Datasets
license: other
task_categories:
  - image-classification
  - object-detection
  - image-segmentation
  - keypoint-detection
  - tabular-classification
  - tabular-regression
  - text-classification
tags:
  - anylearning
  - computer-vision
  - tabular
  - evaluation
---

# AnyLearning datasets

This repository contains reproducible sample datasets used to develop and test
[AnyLearning OSS](https://github.com/nrl-ai/anylearning-oss).

Dataset licenses are recorded in [LICENSES.md](LICENSES.md). The repository's
scripts and original documentation are Apache-2.0, but that license does not
override the terms of any dataset. **Check the dataset license before use.**

### Licence-cleared

| Task | Dataset | Licence |
|---|---|---|
| Image classification | [ZhangLabData: Chest X-Ray](zhanglabdata_chest_xray/) | CC BY 4.0 |
| Object detection | [Safety Helmet and Reflective Jacket Detection](helmet_jacket/) | Apache 2.0 |
| Semantic segmentation | [Electron Microscopy Particle Segmentation](electron_microscopy_particle_segmentation/) | CC BY 4.0 |
| Handpose classification | [American Sign Language Letters](american_sign_language/) | Public Domain |
| Keypoint detection | [Generated Keypoint Stick Figures](keypoint_stick_figures/) | CC0 1.0 |
| Keypoint detection | [Desert Locust](desert_locust_keypoints/) | Apache 2.0 |
| Tabular classification | [UCI Bank Marketing](structured/) | CC BY 4.0 |
| Tabular regression | [UCI Concrete Compressive Strength](structured/) | CC BY 4.0 |
| Text classification | [Banking77](structured/) | CC BY 4.0 |
| Response evaluation | [TruthfulQA](structured/) | Apache 2.0 |

The first four are the real datasets AnyLearning's test suite loads via `REAL_DATASETS` in
`tests/fixtures/datasets.py`.

The keypoint set is a deterministic, generated starter project: 24 training
images, 8 validation images, two instances per image and five named landmarks.
It is intentionally simple and is suitable for tutorials and mechanical tests,
not model-quality claims.

Small, ready-to-upload keypoint archives are also mirrored on the AnyLearning
CDN:

- [CC0 stick figures: train.zip](https://cdn.anylearning.nrl.ai/datasets/keypoint_stick_figures/train.zip)
  and [valid.zip](https://cdn.anylearning.nrl.ai/datasets/keypoint_stick_figures/valid.zip)
- [CC BY 4.0 vertebral X-ray subset: 64-image train.zip](https://cdn.anylearning.nrl.ai/datasets/vertebral_keypoints/v1/vertebral-keypoints-train-64.zip)
  and [16-image valid.zip](https://cdn.anylearning.nrl.ai/datasets/vertebral_keypoints/v1/vertebral-keypoints-valid-16.zip)
- [Apache-2.0 Desert Locust: 630-image train.zip](https://huggingface.co/datasets/nrl-ai/anylearning-data/resolve/main/desert_locust_keypoints/train.zip)
  and [70-image valid.zip](https://huggingface.co/datasets/nrl-ai/anylearning-data/resolve/main/desert_locust_keypoints/valid.zip)

Every vertebral archive embeds its attribution and conversion notes. See the
dataset README before using medical images; this is an engineering example,
not a clinically validated dataset.

The structured-data catalog is available at
[`/datasets/structured/catalog-v1.json`](https://cdn.anylearning.nrl.ai/datasets/structured/catalog-v1.json).
AnyLearning can install its real examples directly from the Dataset workspace;
each entry carries its source, citation, license, size and SHA-256 checksum.

### Fetch on demand for engineering validation

| Task | Dataset | Why it is not committed |
|---|---|---|
| Keypoint detection | [Spondylolisthesis Vertebral Landmark v1](vertebral_keypoints/) | The complete source is fetched from its fixed DOI; a small attributed subset is mirrored on the CDN. |

Datasets without verified redistribution terms are intentionally excluded from
this public repository. See [LICENSES.md](LICENSES.md) for the acceptance policy.

## Using these with AnyLearning

Each dataset ships as per-subset zips. Download this repository next to the
`anylearning-oss` checkout, or point `ANYLEARNING_DATA_DIR` at it, and the test suite
will find it automatically:

```bash
hf download nrl-ai/anylearning-data --repo-type dataset --local-dir ../anylearning-data
pytest tests/e2e/test_real_datasets_e2e.py
```

Archives are extracted into `~/.cache/anylearning-test-data/`, never back into a
repository. See `docs/testing.md` in the main repository.

## Community

Please read [CONTRIBUTING.md](CONTRIBUTING.md) before submitting a dataset and
report vulnerabilities according to [SECURITY.md](SECURITY.md). General support
questions belong in the
[AnyLearning OSS discussions](https://github.com/nrl-ai/anylearning-oss/discussions).
Participation is governed by [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md).