# PhotoAIBench test photos: public domain or CC0 only.
#
# Every photo here is either dedicated to the public domain under CC0 by its author, or in the
# public domain (US federal government work, or published/made long enough ago), as recorded by the
# institution that holds it. No research-only datasets (DIV2K, Flickr2K, FFHQ...), nothing with a
# "non-commercial" or "share-alike" condition, nothing whose status is unclear.
#
# No photos of identifiable living people. The tests upload photos to third-party services and
# publish crops of the results, which we will not do with someone's face without their consent.
# So the portraits are historical: the sitters died decades ago. (A modern digital portrait made
# with a signed release is in properties/photo-ai/BACKLOG.md.)
#
# The image files are not committed. Download and verify them with:
#   python data/photo/fetch_samples.py           (saves to data/private/photo-samples/<id>.<ext>)
#
# Fields
#   id, title (as the holding institution gives it), category (portrait | landscape | texture-text |
#   old-damaged), use (upscale: scored in the upscaler test | restoration: side-by-side only),
#   source_url (the exact file downloaded), page_url (the public record for it), license (exact
#   wording of the license or rights statement as shown on page_url), license_basis (why it is
#   free to use), author, credit (the credit line the holder asks for, where it gives one),
#   retrieved (date downloaded), sha256 and bytes (of the downloaded file), width/height (pixels),
#   format, icc (embedded colour profile, if any), reference (how the upscaler test's reference
#   image is made from the file; see data/photo/bench/prepare.py), notes.
#
# Reference recipe (bench/prepare.py): open the file; apply any EXIF orientation; convert to
# sRGB if a different ICC profile is embedded; crop to `reference.crop` [left, top, right, bottom]
# if given (to cut film borders); resize so the long side is 2000 px with Pillow LANCZOS; trim the
# right/bottom edge to a multiple of 4. The low-resolution input is that reference reduced 4x with
# Pillow BICUBIC.
version: 1
retrieved_by: PhotoAIBench
samples:
  # ------------------------------------------------------------------ portraits
  - id: portrait-little-norway-guide
    title: "Guide at Little Norway, Blue Mounds, Wis."
    category: portrait
    use: upscale
    source_url: https://tile.loc.gov/storage-services/master/pnp/fsac/1a55000/1a55000/1a55005u.tif
    page_url: https://www.loc.gov/item/2017878934/
    license: "No known restrictions on publication."
    license_basis: >-
      Library of Congress rights statement for the collection: "The contents of the Library of
      Congress Farm Security Administration/Office of War Information Color Photographs are in the
      public domain and are free to use and reuse." Made by a US federal agency photographer.
    author: "Arthur Rothstein (1915-1985), photographer, for the US Farm Security Administration"
    credit: "Library of Congress, Prints & Photographs Division, Farm Security Administration/Office of War Information Color Photographs"
    date_made: "August 1942, per the LoC record (Kodachrome transparency; call number LC-USW36-1088)"
    retrieved: 2026-10-07
    sha256: 2f90076e4a273285c5f568081ab026d9039cf879e4faae634240c7988d0e3114
    bytes: 101113476
    width: 5063
    height: 6657
    format: TIFF (8-bit RGB, uncompressed master scan)
    icc: null
    reference: { crop: [304, 266, 4557, 6124], long_side: 2000 }
    notes: >-
      Close head-and-shoulders portrait in colour: skin texture and wrinkles, a striped knitted
      cap, an embroidered vest. Film grain at full resolution, which the 2000 px reference averages
      away. The crop removes the transparency's black film border and edge markings.

  - id: portrait-garage-mechanic
    title: "Garage mechanic near Newark, N.J. Badge denotes member of Office of Defense Transportation"
    category: portrait
    use: upscale
    source_url: https://tile.loc.gov/storage-services/master/pnp/fsac/1a34000/1a34500/1a34582u.tif
    page_url: https://www.loc.gov/item/2017878857/
    license: "No known restrictions on publication."
    license_basis: >-
      Library of Congress rights statement for the collection: "The contents of the Library of
      Congress Farm Security Administration/Office of War Information Color Photographs are in the
      public domain and are free to use and reuse." Made by a US federal agency photographer.
    author: "Marjory Collins (1912-1985), photographer, for the US Office of War Information"
    credit: "Library of Congress, Prints & Photographs Division, Farm Security Administration/Office of War Information Color Photographs"
    date_made: "December 1943, per the LoC record (Kodachrome transparency; call number LC-USW36-1064)"
    retrieved: 2026-10-07
    sha256: 8f3b250fb66bd942bfa360a76a6c1e366ee5d86140ad17708018519612aecb8d
    bytes: 147553030
    width: 6162
    height: 7981
    format: TIFF (8-bit RGB, uncompressed master scan)
    icc: null
    reference: { crop: [555, 399, 5607, 7422], long_side: 2000 }
    notes: >-
      Three-quarter-length colour portrait with a plaid shirt (fine repeating pattern), a small
      badge, and painted signs ("PARTS", "SUPER") behind the sitter: faces, pattern and lettering
      in one frame. The crop removes the film border.

  - id: portrait-walter-white
    title: "Walter Francis White by Clara Sipprell"
    category: portrait
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/5/5a/Walter_Francis_White_by_Clara_Sipprell.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Walter_Francis_White_by_Clara_Sipprell.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: >-
      Released under CC0 by the Smithsonian National Portrait Gallery (object NPG.82.197), as
      recorded on the Commons file page.
    author: "Clara Sipprell (photographer, circa 1950); digital restoration by Adam Cuerden"
    credit: "National Portrait Gallery, Smithsonian Institution (NPG.82.197)"
    date_made: "circa 1950"
    retrieved: 2026-10-07
    sha256: 5d13f451183c2c265cb44525a5e0e5d64fca3d35a260e008754a448f63bbd639
    bytes: 6061381
    width: 3520
    height: 4536
    format: JPEG
    icc: null
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Monochrome studio portrait (warm-toned print), soft window light, dark suit against a dark
      background: low contrast detail in shadows, hair and hands. The Commons file is a cleaned-up
      scan (dust removed), not a raw scan.

  - id: portrait-graham-bell
    title: "Alexander Graham Bell 1895 NPG 77 363"
    category: portrait
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/9/9c/Alexander_Graham_Bell_1895_NPG_77_363.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Alexander_Graham_Bell_1895_NPG_77_363.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: >-
      Released under CC0 by the Smithsonian National Portrait Gallery (Smithsonian Open Access),
      as recorded on the Commons file page.
    author: "Unknown photographer"
    credit: "National Portrait Gallery, Smithsonian Institution (NPG.77.363)"
    date_made: "circa 1895"
    retrieved: 2026-10-07
    sha256: d914c81f1c78bb598bedea4e45ae0ac10eae5729afa18aedf0c77e1e832aa272
    bytes: 39919879
    width: 4726
    height: 7001
    format: JPEG
    icc: Adobe RGB (1998) (converted to sRGB when the reference is made)
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Sepia albumen-era portrait: a full white beard and hair (fine, high-contrast strands), a
      pinstriped jacket, a plain backdrop. Tests whether an upscaler invents hair texture.

  # ------------------------------------------------------------------ landscapes
  - id: landscape-jacques-cartier
    title: "Parc national de la Jacques-Cartier, Quebec, Canada 22"
    category: landscape
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/e/e8/Parc_national_de_la_Jacques-Cartier%2C_Quebec%2C_Canada_22.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Parc_national_de_la_Jacques-Cartier,_Quebec,_Canada_22.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Wilfredor"
    credit: "Wilfredor, Wikimedia Commons, CC0"
    date_made: 2022-08-20
    retrieved: 2026-10-07
    sha256: ea02b50c4acbb2811595c9befdd550fc7f40ec07be8e5581a74606d515aa467b
    bytes: 53962529
    width: 8256
    height: 5504
    format: JPEG
    icc: sRGB IEC61966-2.1
    reference: { crop: null, long_side: 2000 }
    notes: >-
      River with rocks, a forested mountain and sky: foliage (dense, irregular texture that
      upscalers tend to smear or over-sharpen), water ripples, clouds.

  - id: landscape-mount-ida
    title: "Mount Ida chain Messara plain from Phaistos Crete Greece"
    category: landscape
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/6/6d/Mount_Ida_chain_Messara_plain_from_Phaistos_Crete_Greece.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Mount_Ida_chain_Messara_plain_from_Phaistos_Crete_Greece.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Jebulon"
    credit: "Jebulon, Wikimedia Commons, CC0"
    date_made: 2015-02-25
    retrieved: 2026-10-07
    sha256: 9b99e2cded5c12ddfdb760db2e6423708f81f6bd9ecea07c70e8924daf4b683c
    bytes: 13385496
    width: 4479
    height: 2730
    format: JPEG
    icc: sRGB IEC61966-2.1
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Wide view over farmland and olive groves to snow-capped mountains: thousands of tiny trees,
      field edges and distant buildings, the kind of fine detail that disappears at 4x reduction.

  - id: landscape-rio-skyline
    title: "Rio de Janeiro skyline and Sugarloaf Mountain at sunset, Brazil 3"
    category: landscape
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/0/04/Rio_de_Janeiro_skyline_and_Sugarloaf_Mountain_at_sunset%2C_Brazil_3.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Rio_de_Janeiro_skyline_and_Sugarloaf_Mountain_at_sunset,_Brazil_3.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Wilfredor"
    credit: "Wilfredor, Wikimedia Commons, CC0"
    date_made: 2023-06-03
    retrieved: 2026-10-07
    sha256: c3b18c9ff8b5f381e07ca6abd1cc62b4691ffa50afafa60a95821c07ac94832b
    bytes: 17597171
    width: 8043
    height: 4919
    format: JPEG
    icc: ProPhoto RGB (converted to sRGB when the reference is made)
    reference: { crop: null, long_side: 2000 }
    notes: >-
      City, mountains and a bay full of moored boats in haze: straight building edges, many small
      windows, tiny boats. The file is tagged ProPhoto RGB; tools that ignore the profile would
      see desaturated colours, so the reference is converted to sRGB first.

  - id: landscape-sunset-key
    title: "Sunset Key and Sailboats, Florida, 2025"
    category: landscape
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/7/7a/Sunset_Key_and_Sailboats%2C_Florida%2C_2025.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Sunset_Key_and_Sailboats,_Florida,_2025.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Julian Lupyan"
    credit: "Julian Lupyan, Wikimedia Commons, CC0"
    date_made: 2025-11-26
    retrieved: 2026-10-07
    sha256: 9804ac2c73f8a478d01f6648b2e9a06a6779f20e96f3a7e1cf1d6c67d1fce2a7
    bytes: 12520519
    width: 6964
    height: 3626
    format: JPEG
    icc: sRGB IEC61966-2.1
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Sea, sky and an island with sailboats: large smooth areas (where upscalers can add
      banding or false texture) and thin masts and rigging (one or two pixels wide after
      reduction).

  # ------------------------------------------------------------------ texture and text
  - id: text-hotel-post-sign
    title: "Hotel Post, Inn sign, Murnau, Bavaria, Germany"
    category: texture-text
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/b/ba/Hotel_Post%2C_Inn_sign%2C_Murnau%2C_Bavaria%2C_Germany.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Hotel_Post,_Inn_sign,_Murnau,_Bavaria,_Germany.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Jebulon"
    credit: "Jebulon, Wikimedia Commons, CC0"
    date_made: 2013-10-20
    retrieved: 2026-10-07
    sha256: d78e3e889fc5bb009bdb554d3173e92d3a2f96e3e53b5ae8ea4fb318678f3ad9
    bytes: 8976819
    width: 2988
    height: 4248
    format: JPEG
    icc: sRGB IEC61966-2.1
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Painted inn sign with blackletter lettering, a gilded coach and wrought iron, flower boxes
      and a facade: lettering whose shapes an upscaler must not change.

  - id: text-posted-sign
    title: "Posted Nature Sanctuary sign, Capen Hill Nature Sanctuary"
    category: texture-text
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/b/b0/Posted_Nature_Sanctuary_sign%2C_Capen_Hill_Nature_Sanctuary.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Posted_Nature_Sanctuary_sign,_Capen_Hill_Nature_Sanctuary.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Peter Cooper Jr."
    credit: "Peter Cooper Jr., Wikimedia Commons, CC0"
    date_made: 2025-01-04
    retrieved: 2026-10-07
    sha256: 92237613811e102f50f4baa4b55e5321b0f9250ad844d8423bf980d1cef7aa93
    bytes: 10454515
    width: 6000
    height: 4000
    format: JPEG (MPO container from the camera; the first image is used)
    icc: null
    reference: { crop: null, long_side: 2000 }
    notes: >-
      An orange printed sign on a tree in a bare winter forest. At 2000 px the small print is a
      few pixels tall, and after 4x reduction it is illegible: an upscaler that "restores" it is
      inventing letters. Bark and branches give fine natural texture.

  - id: texture-lynx
    title: "Portrait of a Carpathian Lynx"
    category: texture-text
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/7/7c/Portrait_of_a_Carpathian_Lynx.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Portrait_of_a_Carpathian_Lynx.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Wilfredor"
    credit: "Wilfredor, Wikimedia Commons, CC0"
    date_made: 2024-02-05
    retrieved: 2026-10-07
    sha256: 144addb9f175602eaea8da9f53ee37b153b62c015015775e29e7ad8cfc3b6e33
    bytes: 15404425
    width: 7094
    height: 4998
    format: JPEG
    icc: ProPhoto RGB (converted to sRGB when the reference is made)
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Fur, whiskers and ear tufts against a blurred background: the classic case where GAN
      upscalers paint plausible but invented fur.

  - id: texture-golden-chapel
    title: "Golden Chapel, Recife, Pernambuco, Brazil"
    category: texture-text
    use: upscale
    source_url: https://upload.wikimedia.org/wikipedia/commons/0/08/Golden_Chapel%2C_Recife%2C_Pernambuco%2C_Brazil.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Golden_Chapel,_Recife,_Pernambuco,_Brazil.jpg
    license: "CC0 (Creative Commons Zero, Public Domain Dedication)"
    license_basis: "Dedicated to the public domain under CC0 by its author on Wikimedia Commons."
    author: "Wilfredor"
    credit: "Wilfredor, Wikimedia Commons, CC0"
    date_made: 2020-01-07
    retrieved: 2026-10-07
    sha256: c2984dbfd643296079f8f4c624fedd3198862bc3c4740fc6788728f2b248b769
    bytes: 57890250
    width: 6473
    height: 6255
    format: JPEG
    icc: sRGB IEC61966-2.1
    reference: { crop: null, long_side: 2000 }
    notes: >-
      Baroque chapel interior covered in gilded carving, painted panels and blue-and-white tiles:
      dense ornament at every scale.

  # ------------------------------------------------------------------ old and damaged (restoration)
  - id: damaged-daguerreotype-copy
    title: "[Unidentified copy of a daguerreotype portrait of a woman]"
    category: old-damaged
    use: restoration
    source_url: https://tile.loc.gov/storage-services/master/pnp/cph/3c10000/3c10000/3c10000/3c10011u.tif
    page_url: https://www.loc.gov/item/2004664223/
    license: "No known restrictions on publication."
    license_basis: >-
      Library of Congress rights statement for the collection: "The contents of the Library of
      Congress Daguerreotypes Collection are in the public domain and are free to use and reuse."
      Made between 1844 and 1860.
    author: "Mathew B. Brady studio (approximately 1823-1896)"
    credit: "Library of Congress, Prints & Photographs Division, Daguerreotypes Collection"
    date_made: "between 1844 and 1860 (sixth-plate daguerreotype; DAG no. 212)"
    retrieved: 2026-10-07
    sha256: 4e1cae5ac41374a0887e7052808a276ae71df1945de73b17a72e66fad1fded6b
    bytes: 13498340
    width: 3295
    height: 4096
    format: TIFF (8-bit greyscale scan of a film copy negative)
    icc: null
    reference: null
    notes: >-
      Heavily damaged: stains, spots, scratches and a cloudy surface over the whole plate. A hard
      restoration case with no "correct" answer, so it is shown side by side, never scored.

  - id: damaged-daguerreotype-man
    title: "[Unidentified man, head-and-shoulders portrait, facing front]"
    category: old-damaged
    use: restoration
    source_url: https://tile.loc.gov/storage-services/master/pnp/cph/3g10000/3g10000/3g10100/3g10108u.tif
    page_url: https://www.loc.gov/item/2004664536/
    license: "No known restrictions on publication."
    license_basis: >-
      Library of Congress rights statement for the collection: "The contents of the Library of
      Congress Daguerreotypes Collection are in the public domain and are free to use and reuse."
      Made circa 1855.
    author: "F. (Francis) Grice, photographer"
    credit: "Library of Congress, Prints & Photographs Division, Daguerreotypes Collection"
    date_made: "circa 1855 (ninth-plate daguerreotype; DAG no. 1341)"
    retrieved: 2026-10-07
    sha256: 589bacec8517527dff61a29a4fe0c01b65d195bc12c19d5acf50bc5a3087cc6f
    bytes: 52792704
    width: 3746
    height: 4697
    format: TIFF (8-bit RGB scan of a colour copy transparency made before conservation)
    icc: null
    reference: null
    notes: >-
      Tarnish, specks and dust on the plate, inside its brass mat and case: the typical condition
      of a family daguerreotype. Side by side only.

  - id: damaged-tintype-hat
    title: "[Half-length portrait of an African American woman wearing a hat and holding parasol]"
    category: old-damaged
    use: restoration
    source_url: https://tile.loc.gov/storage-services/master/pnp/ppmsca/11000/11090u.tif
    page_url: https://www.loc.gov/item/2010650827/
    license: "No known restrictions on publication."
    license_basis: >-
      Library of Congress rights advisory for the item (William Gladstone Collection of African
      American Photographs). Made between 1861 and 1870 by an unidentified photographer, so in the
      public domain by age in the US.
    author: "Unidentified photographer"
    credit: "Library of Congress, Prints & Photographs Division, Gladstone Collection of African American Photographs"
    date_made: "between 1861 and 1870 (tintype; AMB/TIN no. 1334)"
    retrieved: 2026-10-07
    sha256: 4d172b8106fd2473ee7385c67a611bbeee626f45342b7946f91bdb47185148ea
    bytes: 40123051
    width: 3051
    height: 4383
    format: TIFF (8-bit RGB scan)
    icc: null
    reference: null
    notes: >-
      Scratches and corner damage, low contrast, a dark painted backdrop: a realistic family-album
      tintype. Side by side only.

  - id: damaged-cracked-negative
    title: "Colorado River. Grand Canyon, Tapeets Creek. (Note: it appears that the glass negative may have broken or cracked all the way across the middle of the photo...)"
    category: old-damaged
    use: restoration
    source_url: https://upload.wikimedia.org/wikipedia/commons/1/1f/Colorado_River._Grand_Canyon%2C_Tapeets_Creek._%28Note%2C_it_appears_that_the_glass_negative_may_have_broken_or_cracked_all..._-_NARA_-_518020.jpg
    page_url: https://commons.wikimedia.org/wiki/File:Colorado_River._Grand_Canyon,_Tapeets_Creek._(Note,_it_appears_that_the_glass_negative_may_have_broken_or_cracked_all..._-_NARA_-_518020.jpg
    license: "Public domain (work of the US federal government)"
    license_basis: >-
      Made by photographers of the Powell survey of the Colorado River for the US government; held
      by the US National Archives (NAID 518020). Commons records it as public domain, US
      government work.
    author: "Elias Olcott Beaman, James Fennemore or John Karl Hillers (Powell survey photographers)"
    credit: "U.S. National Archives and Records Administration (NAID 518020)"
    date_made: "between 1871 and 1878 (stereograph from a glass negative)"
    retrieved: 2026-10-07
    sha256: 89b71856bc1d09b029b66c0495a6cf8a980b25957703a87cccdc164ecdc515d0
    bytes: 1753075
    width: 3000
    height: 1871
    format: JPEG (greyscale)
    icc: null
    reference: null
    notes: >-
      A stereo pair of canyon walls with a crack running across both halves: the classic "repair
      the crack" restoration task. Side by side only.