RevolutionCrossroads/si_us_revolutionary_era_collections

Dataset Card for Smithsonian American Revolutionary Era Collections

Dataset Summary

A specially selected subset of the Smithsonian’s Open Access collections covering objects from 1770–1810 selected for the Revolution Crossroads project in honor of the 250th anniversary of the founding of the United States. Drawn from four museums—the National Museum of American History, National Postal Museum, Smithsonian American Art Museum, and National Portrait Gallery—the dataset includes 12,667 records with descriptive metadata, 4,209 records with linked images (6,242 images total), and AI-generated OCR for selected images. It provides a foundation for exploring Revolutionary-era material culture and developing multimodal research and analysis tools.

Dataset Description

The Smithsonian Revolutionary Era Collections dataset is a specially selected subset of the Smithsonian’s Open Access corpus, created as part of the Revolution Crossroads project in honor of the 250th anniversary of the United States.

It brings together descriptive records and related images from across four Smithsonian museums—the National Museum of American History, the National Postal Museum, the Smithsonian American Art Museum, and the National Portrait Gallery—that roughly date to the period 1770–1810, the era of the American Revolution and early republic.

The dataset includes: - 12,667 records describing museum objects and archival items - Metadata fields converted from Smithsonian Open Access JSON to Parquet format for analysis - 4,209 records linked to digital surrogates (thumbnail images and full object files) with 6,242 total images

This dataset was assembled to support research and experimentation at the intersection of cultural heritage and artificial intelligence. It provides a structured corpus for exploring American Revolutionary-era material culture, testing multimodal approaches that link descriptive metadata with images, and developing tools for analysis, visualization, and discovery.

Dataset Details

• Prepared by: Smithsonian Institution, Office of Digital & Innovation staff
• Shared by: Revolution Crossroads
• Language(s): English
• License: CC0-1.0

Dataset Sources

• Repository: Smithsonian Open Access, Smithsonian Institution
• Records contributed by:
  • National Museum of American History
  • National Postal Museum
  • Smithsonian American Art Museum
  • National Portrait Gallery
• Access Path: All records were retrieved through the Smithsonian Open Access API

Curation Rationale

This dataset was curated to support the Revolution Crossroads project, which explores the Revolutionary and early republic period in recognition of the 250th anniversary of the United States.

The goal was to assemble a cross-museum dataset that: - Highlights objects dating between 1770–1810, the years spanning the American Revolution and early republic - Brings together holdings from across Smithsonian museums to allow comparative and cross-collection analysis - Focuses on descriptive metadata and images, fields most likely to support research, teaching, and machine learning applications - Provides a manageable, structured corpus for use in AI and digital humanities experimentation, rather than exposing the full complexity of Smithsonian Open Access

Because of these qualities, the dataset lends itself to a range of applications, including: - Exploring late Colonial and early American material culture through museum object metadata and images - Studying connections between objects, people, places, and themes represented in Smithsonian collections - Developing multimodal tools that combine descriptive metadata with linked images - Training and evaluating models for image classification, metadata-based clustering, or information retrieval in cultural heritage contexts - Supporting educational and public history projects that highlight Smithsonian holdings from the Revolutionary era

Dataset Creation

The Smithsonian Revolutionary Era Collections dataset was created by identifying objects and artwork records dated to the period 1770–1810 across four Smithsonian museums, in collaboration with curators and collections staff. Once the target lists of objects were compiled, the corresponding records were retrieved through Smithsonian Open Access.

National Postal Museum Smithsonian American Art Museum National Portrait Gallery - These museums were able to generate relatively complete listings of Revolutionary-era holdings.

National Museum of American History (NMAH) - NMAH records required additional review. - Date fields are free-text and uncontrolled, making searching difficult. - As a result: - Some non-period materials were swept up in initial searches (e.g., objects about the Revolutionary era but created later) - Manual cleanup was used to exclude broad ranges that fell mostly outside of scope (e.g., “1800–1900”) and replicas/commemoratives where identifiable - Some true period objects may still be missing due to incomplete or provisional records

Notes on Completeness

The dataset is not comprehensive. It represents the subset of Revolutionary-era records identifiable at the time of creation, filtered with curatorial input. Future cataloging work may surface additional relevant objects.

Data Collection and Processing

Processing Steps

• Retrieved JSON records from Open Access for each identified object
• Flattened and converted to Parquet format for efficient analysis
• Retained only fields most relevant to anticipated uses (e.g., title, date, object type, place, names, topics, identifiers, media references)
• Preserved mediaCount, mediaURLs, and thumbnail fields to allow linking to images
• Maintained free-text contextual fields (e.g., date, objectType, name) alongside parallel indexed_* fields for normalized values
• To facilitate downstream AI processing, records with one or more associated images were used to generate a companion PDF for each record. These PDFs combine all available images associated with a record into a single document and are available through the hf_pdf_url field. While originally created to support OCR workflows, they also provide a convenient way for researchers to view all images associated with a record in a single file.

Supporting Files Available

JSON-formatted data from Open Access are available in the Files tab:

• "api_json" - contains the JSON output of the records from the dataset, stored in directories for each of the 4 museums included. The file naming convention is "[unit]_[object_number].json", with the object number and unit abbreviation from the records. In the main dataset, the object_number may be found in the "EDANid" fields after "edanndm:".

.jpg format media files retrieved via Open Access are available in the Files tab:

• "media" contains .jpg files for all media from the included records. The file names are unchanged from retrieval and reflect different naming practices of each museum, but generally include the museum acronym, an ID number, and often an appended number or letter if there are multiple images for a single record. In the main dataset, the associated image file names may be found as "id" in the "media_url" fields.

.pdf format companion files generated by the Revolution Crossroads project are available in the Files tab:

• media/pdfs contains one PDF for each record with one or more associated images. Each PDF combines all images associated with a single record into a single document to support downstream AI processing and provide a convenient format for viewing multi-image records. In the main dataset, the corresponding PDF may be accessed through the hf_pdf_url field.

Accessing a Specific Data Version

This dataset may be updated over time as source data is revised, normalized, or restructured. If you are using the data for a project, we recommend accessing a specific tagged version so your workflow remains stable and reproducible.

You can access a tagged version using the revision parameter in load_dataset:

from datasets import load_dataset
import polars as pl

dataset_name = "RevolutionCrossroads/si_us_revolutionary_era_collections"
commit_tag = "Source_2026-06-09"

ds = load_dataset(
    dataset_name,
    revision=commit_tag,
    split="train"
)

# Convert to pandas
pandas_df = ds.data.table.to_pandas()

# Convert to Polars
polars_df = pl.from_arrow(ds.data.table)

This example loads the specified tagged dataset version and converts it to either a pandas or Polars dataframe for analysis.

Available Data Versions

• 2025-09-24 – Initial public release.

• 2025-11-25 – Updated source data.

• Source_2026-06-09 – Updated source data and revised dataset structure.

• new_OCR_v1 – Adds AI-generated Chandra OCR 2 (RevX_OCR_text) for selected images while preserving the dataset structure.

Using a tagged version is recommended for research, publication, or project workflows where consistency over time is important.

AI OCR Enrichment

This dataset has been enriched with AI-generated OCR produced using DataLab's Chandra OCR 2, a Vision Language Model (VLM) for document understanding.

Complete Chandra OCR 2 processing outputs are stored in the Revolution Crossroads VLM OCR Hugging Face bucket. The bucket includes a README describing the processing pipeline, output formats, and directory structure.

• Complete VLM OCR Bucket: https://huggingface.co/buckets/RevolutionCrossroads/RevolutionCrossroads_VLM_OCR
• Smithsonian Collections: https://huggingface.co/buckets/RevolutionCrossroads/RevolutionCrossroads_VLM_OCR/tree/smithsonian

One additional field has been added:

• RevX_OCR_text contains plain text extracted from selected Smithsonian collection images using Chandra OCR 2.

Unlike the newspaper and pension datasets, OCR was not generated for every image. An initial review was used to identify images likely to contain readable textual content, and only those images were processed. Images that do not contain substantial text, or that were not selected for processing, have null values in this field.

These outputs represent the direct results of automated AI processing. They have not been manually validated or corrected and should be treated as machine-generated data.

A subsequent Smithsonian data refresh added additional records and images. These newly available images will be included in a future round of Chandra OCR 2 processing and incorporated into subsequent dataset releases.

Quality Considerations

• The dataset was not normalized or corrected
• Some non-period records may remain, and some relevant ones may be missing
• Metadata inconsistencies reflect differences in cataloging practice across museums

Dataset Structure

The dataset consists of two main components. Descriptive Records - 12,667 records representing museum objects and archival items - Originally in JSON via Smithsonian Open Access, converted to Parquet for analysis and ease of loading - Each record corresponds to a single object/item - Includes Smithsonian metadata fields such as title, description, date, place, subject terms, medium, dimensions, and credit line - Each record has a catalog number or identifier (not guaranteed unique)

Linked Images - 4,209 records include digital surrogates (thumbnail images and links to full object files), representing 6,242 total images - Images are referenced by URL and can be retrieved directly from Smithsonian servers - Image file names and URLs are tied to their parent object record via identifier

File Organization

• Parquet files contain the flattened descriptive metadata.
• Associated JPEG image files are available in the dataset Files tab and are referenced from the metadata through the mediaURLs field.
• Companion PDFs generated by the Revolution Crossroads project are available through the hf_pdf_url field. These PDFs combine all images associated with a record into a single document to support downstream processing and provide a convenient way to view multi-image records.
• The dataset can be loaded directly into a dataframe-style environment for analysis, with links to associated media preserved in the metadata.

Data Fields

Core Descriptive Fields

collectionsURL (string)
Public Smithsonian Collections record URL.
Example: https://collections.si.edu/search/detail/edanmdm:saam_XX108A

unitCode (string)
Code for the contributing Smithsonian museum or unit.
Example: SAAM

dataSource (string)
Human-readable name of the contributing museum or unit.
Example: Smithsonian American Art Museum

title (string)
Cataloged title of the object.
Example: George Washington

EDANid (string)
Smithsonian Enterprise Digital Asset Network (EDAN) record identifier.
Example: edanmdm:saam_XX108A

guid (string)
Persistent Smithsonian GUID/ARK resolver URL.
Example: http://n2t.net/ark:/65665/vk7c06aa8a1-8da1-47c8-884f-6a47cc2e6305

recordLink (string)
URL to the object record on the contributing unit's website, when available. Users may use guid as the persistent record link when this field is empty.
Example: https://americanart.si.edu/collections/search/artwork/?id=27849

lastUpdateDate (date32)
Date the source record was last updated in Smithsonian Open Access.
Example: 2025-09-05

Descriptive Content

creditLine (string)
Credit or source information associated with the object.
Example: ["Smithsonian American Art Museum, Transfer from the U.S. Department of the Interior through the General Services Administration"]

date (string)
Date or date range associated with the object.
Example: ["ca. 1800"]

identifier (string)
Museum accession number, catalog number, or other object identifier. These values are not guaranteed to be unique across contributing units.
Example: ["XX108A"]

name (string)
Names of creators, artists, sitters, and other associated individuals, including their contextual roles when available.
Example: {"Attributed to": "William Winstanley, born England, active 1775-1806", "Sitter": "George Washington"}

notes (string)
Additional descriptive, historical, exhibition, or catalog notes associated with the object. Values may include contextual labels.

physicalDescription (string)
Information about the object's medium, materials, dimensions, and other physical characteristics.
Example: {"Medium": "oil on canvas", "Dimensions": "30 1/2 x 25 1/4 in. (77.4 x 64.2 cm.)"}

place (string)
Place information associated with the object, including contextual labels when available.

publisher (string)
Publisher information for printed works, when available.

alternativeTitles (string)
Alternative or variant titles associated with the object, when available.

objectType (list[string])
Object categories or classifications associated with the record.
Example: ["Painting"]

topic (list[string])
Subject terms or topical descriptions associated with the record.
Example: ["Occupation/political/president", "Portrait male"]

Indexed Fields (normalized values)

Normalized versions of selected descriptive fields. These fields generally omit contextual labels and provide values intended to support searching, filtering, and analysis.

indexed_dates (list[string])
Indexed date periods associated with the object.
Example: ["1800s"]

indexed_names (list[string])
Indexed names of people or organizations associated with the object.
Example: ["Washington, George", "Winstanley, William"]

indexed_object_types (list[string])
Indexed object classifications.
Example: ["Paintings"]

indexed_places (list[string])
Indexed geographic terms associated with the object.
Example: ["Philadelphia", "Pennsylvania"]

indexed_topics (list[string])
Indexed subject and topical terms associated with the object.
Example: ["Occupations", "Men", "Portraits", "Political", "Presidents"]

Media Fields

mediaCount (int32)
Number of media items associated with the record.
Example: 1

mediaURLs (list[string])
URLs for digital media associated with the record.
Example:
["https://ids.si.edu/ids/deliveryService?id=SAAM-XX108A_1"]

thumbnail (string)
URL for a preview thumbnail image, when available.
Example:
https://ids.si.edu/ids/iiif/SAAM-XX108A_1/full/200,/0/default.jpg

hf_pdf_url (string)
URL to a Revolution Crossroads-generated PDF prepared from the images associated with the record for downstream processing and analysis.
Example:
https://huggingface.co/datasets/RevolutionCrossroads/si_us_revolutionary_era_collections/resolve/main/media/pdfs/saam_XX108A.pdf

RevX_OCR_text (string)
AI-generated OCR produced from selected images using DataLab Chandra OCR 2. Values correspond to the images associated with the record.

• [] indicates that the record has no associated images.
• [None] or [""] indicates that images are present but no OCR text was produced.
• Otherwise, the value contains OCR text for the selected processed image or images. Example: ["Extracted text from the image..."]

Source Data

Original Data Producers

Records were created and are maintained by staff at: - National Museum of American History (NMAH) - National Postal Museum (NPM) - Smithsonian American Art Museum (SAAM) - National Portrait Gallery (NPG)

Data Access

• Records were accessed via Smithsonian Open Access after the object lists were confirmed with museum staff
• Open Access provided a standardized way to pull complete records across multiple units

Characteristics of Source Records

• Metadata reflects cataloging practices over many decades and may have changed over time
• Some fields are structured (e.g., accession numbers), while others are free-text with contextual labels (e.g., “date made: circa 1775”)
• Cataloging depth varies by museum and by object: some records are richly detailed, others more minimal

Personal and Sensitive Information

There is no known personal or sensitive information included in this dataset.

Considerations for Using the Data

Risks and Limitations

This dataset represents a curated subset of Smithsonian collections rather than a complete record of American Revolutionary-era holdings. - Some Revolutionary-era objects may not be included if their catalog records lacked sufficient detail or were not discoverable through metadata searches - Some objects outside the 1770–1810 period may be present if their records used broad or ambiguous date ranges (e.g., “1800–1900”) - Metadata fields are descriptive but not always comprehensive. Some records may be incomplete if they have not yet been fully researched or updated

The original records in Smithsonian Open Access were created over many decades by different units and staff, leading to variation in terminology, structure, and level of detail. Cataloging is an ongoing process, and records are continually updated as new research is conducted.

For additional reuse limitations, users should consult the Smithsonian Open Access FAQ and Smithsonian Terms of Use.

Recommendations

Users should be aware of the risks and limitations of the dataset. Additional guidance may be added as further recommendations are developed.

Users should distinguish between descriptive metadata and AI-generated OCR (RevX_OCR_text) when conducting downstream analysis. OCR was produced only for selected images likely to contain meaningful text and has not been manually validated or corrected. In this field, [] indicates that the record has no associated images, while [None] or [""] indicates that images are present but no OCR text was produced.

Additional Information

Citation Information

BibTeX @dataset{RevolutionCrossroads_Smithsonian_2025, author = {Revolution Crossroads Project Team}, title = {Smithsonian American Revolutionary Era Collections}, year = {2025}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/RevolutionCrossroads/si_us_revolutionary_era_collections}, doi = {10.57967/hf/6527} }

APA Revolution Crossroads Project Team. (2025). Smithsonian American Revolutionary Era Collections [Data set]. Hugging Face. https://doi.org/10.57967/hf/6527

Glossary

• EDAN (Enterprise Digital Asset Network): Smithsonian’s internal collections database system. Each object or item has an EDAN identifier
• Indexed fields: Normalized versions of descriptive fields (e.g., indexed_names, indexed_places). These lack contextual labels (like “Maker” or “Sitter”) and provide just the standardized value, making them easier for search and filtering
• Open Access: Smithsonian’s public data service that provides standardized access to digitized collections records and media
• Thumbnail: A small preview image linked to an object, often used to quickly display collections online
• Parquet: A columnar data file format optimized for efficient analysis and storage. Used here to make large metadata files easier to work with

Dataset Card Contact

revolutioncrossroads@si.edu