---
dataset_info:
  splits:
  - name: train
    num_examples: 39353
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
language:
- da
- en
pretty_name: SMK Image-Text (Danish/English)
size_categories:
- 10K<n<100K
license: other
task_categories:
- image-to-text
- feature-extraction
task_ids:
- image-captioning
---

# Dataset Card for “SMK Image-Text (Danish/English)”

## Dataset Description

- **Source:** Statens Museum for Kunst (SMK) collection API.
- **Records:** **39,353** objects with paired images and bilingual metadata.
- **Storage:** 2 Parquet shards on the Hub (`data/train-*.parquet`), ~3 GB total each.
- **Languages:** Danish (`da`) and English (`en`) fields where available.

### Summary

Each row corresponds to an SMK collection object. It contains:
- Raw image bytes (`image_bytes`) plus thumbnails and basic image stats (width/height/size, entropy, contrast, etc.).
- Object metadata in Danish and English: titles, object names, artists/creators, production dates, techniques, materials, inscriptions, labels, documentation references.
- Rights information: `public_domain` flag and `rights` text per object.

---

## Dataset Structure

### Key Fields

- **Images**
  - `image_bytes` (`binary`): full-resolution image bytes. Cast to `datasets.Image()` to decode.
  - `image_thumbnail` (`string` URL), `image_width`, `image_height`, `image_size`, `image_orientation`, `image_cropped`, `colors`, `suggested_bg_color`, `entropy`, `contrast`, `brightness`, `saturation`, `colortemp`.
- **Identity**
  - `object_number`, `id`, `object_url`, `frontend_url`, `responsible_department`.
- **Dating & dimensions**
  - `acquisition_date`, `acquisition_date_precision`, `production_date_en/da` (list with fields `start`, `end`, `start_prec`, `end_prec`, `period`), `dimensions` (list with fields`value`, `unit`, `part`, `type`, `notes`, `precision`).
- **Titles & names**
  - `titles_en/da` (list with fields `language`, `title`, `type`, `notes`, `translation`).
  - `object_names_en/da` (list with fields `name`, `classification_notes`).
- **Creators**
  - `artist_en/da` (list of strings).
  - `production_en/da` (list with fields `creator`, `creator_forename`, `creator_surname`, `creator_gender`, `creator_nationality`, `creator_role`, `creator_history`, `creator_lref`, `creator_qualifier`, `craftsman`, birth/death dates, notes).
- **Techniques & materials**
  - `techniques_en/da`, `materials_en/da`, `medium_en/da`.
- **Context & documentation**
  - `labels_en/da` (list with fields `date`, `source`, `text`, `type`).
  - `inscriptions_en/da` (list with fields `content`, `description`, `language`, `type`, `date`, `notes`).
  - `documentation_en/da` (list with fields `author`, `title`, `shelfmark`, `page_reference`, `year_of_publication`, `notes`).
  - `content_description_en/da`, `production_dates_notes_en/da`.
- **Rights**
  - `public_domain` (bool) and `rights` (string) per object.

---

## Usage

```python
from datasets import load_dataset, Image

ds = load_dataset("V4ldeLund/SMK-image-text", split="train")

# Decode image bytes into PIL images
ds = ds.cast_column("image_bytes", Image())
sample = ds[0]
sample["image_bytes"].show()

# Example: English title and production info
print(sample["titles_en"], sample["production_en"])
```

---


## Intended Uses

- Image captioning, image text retrieval, metadata completion, museum collection exploration, multilingual vision language modelling.

---

## Maintainer / Contact

- **Maintainer:** Vladimir Salnikov — v4ldesalnikov@gmail.com
- **Issues & questions:** Please open a discussion on the dataset’s Hugging Face page: `https://huggingface.co/datasets/V4ldeLund/SMK-image-text`