# Demo pages

These eight pages are a compact, presentation-oriented input set. They are
not ground truth, a benchmark, or evidence of recognition accuracy.

## Layout

Pages are grouped one directory per work, so each directory is directly usable as a
document input while `catalog.csv` still addresses every page individually:

```text
data/demo/
  catalog.csv                 the flat per-page catalog (ids demo-001 .. demo-008)
  beethoven-op026-01/         page-000.png  page-001.png    consecutive pages, one work
  beethoven-op090-01/         page-000.png
  beethoven-op101-01/         page-000.png
  schubert-d911-01/           page-000.png
  schubert-d911-05/           page-000.png
  wagner-wwv086a/             page-000.png  page-002.png    title page + a score page
```

File names carry only the source page number; the directory carries the work, and
`catalog.csv`'s `original_name` column preserves each file's name in the source collection.

## Running it

One page:

```bash
omrpipe run data/demo/beethoven-op026-01/page-000.png -o runs/beethoven-demo
```

The complete sample set, page by page:

```bash
omrpipe matrix data/demo -o runs/demo-matrix
```

A whole work as one document, merged into a single `document.mei`:

```bash
omrpipe document run data/demo/beethoven-op026-01 -o runs/beethoven-document
```

`beethoven-op026-01` is the only multi-page group whose pages are consecutive, so it is the
one to reach for when demonstrating document assembly. Note that assembly is not *joining*: a
movement crossing the page break appears as two adjacent `<mdiv>`s until Stage C lands — see
[`../../docs/document-workflow.md`](../../docs/document-workflow.md).

**`wagner-wwv086a/` currently produces an empty `document.mei`.** Both of its pages are
classified `non_music` and skipped, including `page-002`, which `omrpipe run` transcribes
successfully on its own. Document classification runs before the compatibility upscale, and
these are the smallest rasters in the set (596x842). It is kept as a regression case, not as a
working example.

The current per-page outcomes, including renders of the published-but-degraded MEIs, are documented in
[`../../docs/demo-expected-behavior.md`](../../docs/demo-expected-behavior.md).

## Provenance and rights

The images were selected, without annotations, from the *Bounding Box
Annotation of Musical Measures in Sheet Music* collection by Zalkow, Villar
Corrales, Tsai, Arifi-Mueller, and Mueller (ISMIR 2019). The collection's
README identifies the Beethoven, Schubert, and Wagner page images as public
domain scans from IMSLP. The collection's separately supplied Carus material
is not included. Its measure annotations are licensed separately
(CC BY-NC-SA 4.0) and are **not** included here.

Collection: https://www.audiolabs-erlangen.de/resources/MIR/2019-ISMIR-LBD-Measures

| ID | Source work | Page | SHA-256 |
| --- | --- | --- | --- |
| `demo-001` | Beethoven, Op. 26 No. 1 | 000 | `2d415af22de8a5b2f980d8c73302868581c43f18a190e5169a22a2efc6937126` |
| `demo-002` | Schubert, D. 911 No. 1 | 000 | `995f42a57adeaa376ec6f8964c107611164926f2794ba762ba5600a3831e0609` |
| `demo-003` | Wagner, WWV 86A | 000 | `b2c9fc351a466b00ec0fc21289a62222089c2dd734c61add3ee87aa7e85942be` |
| `demo-004` | Beethoven, Op. 26 No. 1 | 001 | `16d3bd2282810c417a2299aaae163f0ba6035be3f707ccf2865f5a61d02220a5` |
| `demo-005` | Beethoven, Op. 90 No. 1 | 000 | `15fafcf63bc5147c99680d66db754ed66e2f737835fb1952ed2a0c33dd888276` |
| `demo-006` | Beethoven, Op. 101 No. 1 | 000 | `48ecda84debf07fafa67f1131ca3d01bd288bd711bd41f647f957003c2f5eb60` |
| `demo-007` | Schubert, D. 911 No. 5 | 000 | `ac4412efcde935896a34bd86bf7f0b158d9fafef6d2596a2fca62288c41bfb99` |
| `demo-008` | Wagner, WWV 86A | 002 | `141b35c0f00a9a268eb6f0049c7ff2f246f7099ab075a471cfa31e7e07ddbe9f` |

`demo-003` deliberately retains a non-music title page for document-classification demonstrations;
`demo-008` is the first score page from the same Wagner source.