---
pretty_name: SeaShapes
license:
- cc-by-4.0
- apache-2.0
tags:
- 3d
- glb
- maritime
- ships
- synthetic-data
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - "*/preview.png"
  - metadata.jsonl
---

# SeaShapes

3D models of maritime objects, each with its real-world size and its source. One folder per asset, holding the model and a preview. `metadata.jsonl` describes every asset, one row each, and `schema.json` defines the fields.

## Query

In the Data Studio SQL console, or anywhere DuckDB runs:

```sql
SELECT id, category, length_m, width_m, licence
FROM 'hf://datasets/SEA-AI/seashapes/metadata.jsonl'
WHERE list_contains(tags, 'containers') AND length_m BETWEEN 80 AND 100;
```

Pin a file by commit, never by branch:

```
https://huggingface.co/datasets/SEA-AI/seashapes/resolve/<commit>/<id>/<id>.glb
```

## Fields

| Field | Meaning |
|---|---|
| `id` | Folder name |
| `file_name` | Preview image, `<id>/preview.png` |
| `title`, `author`, `source_url`, `licence` | Attribution; `licence` is an SPDX identifier |
| `category`, `tags`, `description` | What the object is; `tags` are free filters |
| `length_m`, `width_m`, `height_m` | Real overall size, keel to top for a vessel |
| `size_source` | How the real size is known; model units are arbitrary |
| `file`, `format`, `size_bytes`, `sha256` | The file |
| `front_axis` | Where the front points in the file; omitted when it is glTF's `+z` |
| `triangles`, `textures`, `texture_max_px` | Mesh detail |
| `modified` | Whether the file differs from what the source offers |
| `added` | Date added |

`schema.json` is the authority; a row that fails it is not added.

## Adding an asset

No clone needed: fetch the small files, add the asset, upload only what changed.

```bash
hf download SEA-AI/seashapes --repo-type dataset --local-dir seashapes --exclude "*.glb" --exclude "*.png"
cd seashapes
```

1. Put the model at `<id>/<id>.glb` and append its row to `metadata.jsonl`. Leave `sha256` and `size_bytes` out; the check fills them in.
2. Render its preview: `uv run tools/preview.py <id>`.
3. Check every row: `uv run tools/validate.py`. Rows whose folder is here are checked against their files, the rest against `schema.json`.
4. Upload in one commit:

```bash
hf upload SEA-AI/seashapes . . --repo-type dataset --include "<id>/*" --include metadata.jsonl --commit-message "Add <id>"
```

Pin the new commit, never `main`, wherever the asset is used.

## Licences

| Files | Licence |
|---|---|
| `<id>/<id>.glb` | The author's, given per row in `metadata.jsonl` |
| `<id>/preview.png` | The model's licence and attribution; each is a render of the model, modified: rendered by SEA.AI |
| `metadata.jsonl`, `schema.json`, `README.md` | CC BY 4.0, SEA.AI |
| `tools/` | Apache-2.0, SEA.AI |

Credit the author and link the source when you share a model or work made from it. Every row of `metadata.jsonl` carries its `title`, `author`, `source_url` and `licence`, and models from Sketchfab embed the same in their glTF `asset.extras`:

```sql
SELECT title, author, source_url, licence
FROM 'hf://datasets/SEA-AI/seashapes/metadata.jsonl';
```

Models from Sketchfab are the GLBs it offers for download, unmodified. Their authors do not endorse SEA.AI or this dataset.

All files are provided as is, without warranties of any kind, to the extent the law allows; see each licence.

Licensing reviewed 2026-10-10.