External/SnapOtter
1
0
Fork 0
mirror of https://github.com/snapotter-hq/SnapOtter.git synced 2026-08-03 07:46:42 +02:00
Code Issues Packages Projects Releases Wiki Activity
Files
5cc0a850c68e0af7a427b8442c24c55f6157416e
SnapOtter/tests/fixtures/ocr-real/manifest.json
T
SnapOtterandGitHub 991c981529 fix: make OCR portable and reliable across AMD64 and ARM64 (#519) 
* fix: make OCR portable and reliable

* fix: harden OCR installation portability

* fix: pin OCR partials across downloads

* fix: make OCR execution reliably asynchronous

* fix: harden OCR portability and docs routes

* fix: preserve decoder and docs safeguards
2026-07-15 03:34:24 +08:00

623 lines
29 KiB
JSON
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
{
"binaryBudgetBytes": 5242880,
"boardCohort": {
"distinctRowGroups": [0, 1, 3],
"fixtureIds": ["jawildtext-board-0001", "jawildtext-board-0049", "jawildtext-board-0127"],
"perImageFastFloor": {
"minimumTokenF1": 0.32,
"minimumTokenPrecision": 0.5,
"minimumTokenRecall": 0.25
},
"selectionRule": "The smallest byte-for-byte source image in each of row groups 0, 1, and 3; image-only visual/privacy review confirmed distinct conditions and safe public content.",
"selectionStatus": "FROZEN_BEFORE_ANY_OCR_OUTPUT",
"stopPolicy": "If any frozen cohort image fails after the general development-corpus fix, stop rotating fixtures and reconsider the Fast CJK architecture."
},
"description": "Pinned, redistributable real-capture OCR regression fixtures with verbatim source annotations.",
"excludedCandidates": [
{
"dataset": "TextOCR",
"reason": "Not redistributed: Open Images says image licenses must be verified per original Flickr image, and that chain of title was not revalidated for a selected TextOCR record.",
"sourceUrls": [
"https://textvqa.org/textocr/dataset/",
"https://github.com/openimages/dataset/blob/master/READMEV3.md"
]
}
],
"fixtures": [
{
"annotation": {
"bytes": 19077,
"path": "annotations/jawildtext-board-0001.json",
"sha256": "4e01a2057c5f2889d78bd26b369b44e53dad9ffcdc69b2a70f28844068cafefc",
"sourceField": "record.polygons"
},
"category": "board-or-sign",
"evaluation": {
"ignoredTokens": ["<UNK>"],
"mode": "annotation-token-coverage",
"notes": "Polygon order is annotation order, not guaranteed page reading order; score per-polygon text coverage and ignore only literal <UNK> labels."
},
"groundTruth": {
"bytes": 1341,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/jawildtext-board-0001.txt",
"sha256": "a8e8a7adef665d299f50a355b5c79d9d2a69e4d0f187fdedfe6312befe446c02",
"sourceField": "record.polygons[*].text",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "jawildtext-board-0001",
"image": {
"bytes": 93653,
"height": 1080,
"mediaType": "image/jpeg",
"path": "images/jawildtext-board-0001.jpg",
"sha256": "04b24d7189f1d905b40ab784ac912c158e2f49377dbf0ec53421b4709a356bfe",
"sourceSha256": "04b24d7189f1d905b40ab784ac912c158e2f49377dbf0ec53421b4709a356bfe",
"width": 1080
},
"language": "ja",
"provenance": {
"annotationLicense": "Apache-2.0",
"attribution": "JaWildText by Koki Maeda and Naoaki Okazaki (LLM-jp), released as an ICDAR 2026 benchmark.",
"dataset": "llm-jp/jawildtext",
"licenseEvidenceUrls": [
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/LICENSE",
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The source row excluding image bytes was serialized as canonical UTF-8 JSON; annotation values and polygon coordinates were not changed.",
"Ground truth lists each polygons[*].text value verbatim in source array order, separated by LF, with one terminal LF."
],
"pixelLicense": "Apache-2.0",
"privacyReview": "Public consumer-product instruction placard; no identifiable person or private data is visible.",
"repositoryRevision": "627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceDatasetUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/tree/627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceRecordUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/board_vqa/train-00000-of-00006.parquet",
"sourceShard": {
"bytes": 558557296,
"globalRowIndex": 0,
"path": "board_vqa/train-00000-of-00006.parquet",
"rowGroup": 0,
"rowIndex": 0,
"sha256": "2a3ebcb8fa627d2477b8fd0190b34f4f41bb235c268131a564ccae57e44d4b14"
},
"split": "train",
"upstreamId": "0001"
}
},
{
"annotation": {
"bytes": 13942,
"path": "annotations/jawildtext-board-0049.json",
"sha256": "f16b1b9e83e99b29ad284aa7e465319eb348fb81637e679a1ed526480d477af4",
"sourceField": "record.polygons"
},
"category": "board-or-sign",
"evaluation": {
"ignoredTokens": ["<UNK>"],
"mode": "annotation-token-coverage",
"notes": "Polygon order is annotation order, not guaranteed page reading order; score per-polygon text coverage and ignore only literal <UNK> labels."
},
"groundTruth": {
"bytes": 1476,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/jawildtext-board-0049.txt",
"sha256": "b745aa777473596f638b7caeff48a8149494f17c5cb0be181b23a63f2a55c8f7",
"sourceField": "record.polygons[*].text",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "jawildtext-board-0049",
"image": {
"bytes": 1071604,
"height": 3338,
"mediaType": "image/jpeg",
"path": "images/jawildtext-board-0049.jpg",
"sha256": "17a3196219f3c512520b3880a36461a771c4558715989696f949a137e6e0867e",
"sourceSha256": "17a3196219f3c512520b3880a36461a771c4558715989696f949a137e6e0867e",
"width": 2160
},
"language": "ja",
"provenance": {
"annotationLicense": "Apache-2.0",
"attribution": "JaWildText by Koki Maeda and Naoaki Okazaki (LLM-jp), released as an ICDAR 2026 benchmark.",
"dataset": "llm-jp/jawildtext",
"licenseEvidenceUrls": [
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/LICENSE",
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The source row excluding image bytes was serialized as canonical UTF-8 JSON; annotation values and polygon coordinates were not changed.",
"Ground truth lists each polygons[*].text value verbatim in source array order, separated by LF, with one terminal LF."
],
"pixelLicense": "Apache-2.0",
"privacyReview": "Public apartment water-service interruption notice; only public contractor phone numbers are visible, with no resident or private data.",
"repositoryRevision": "627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceDatasetUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/tree/627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceRecordUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/board_vqa/train-00000-of-00006.parquet",
"sourceShard": {
"bytes": 558557296,
"globalRowIndex": 57,
"path": "board_vqa/train-00000-of-00006.parquet",
"rowGroup": 1,
"rowIndex": 19,
"sha256": "2a3ebcb8fa627d2477b8fd0190b34f4f41bb235c268131a564ccae57e44d4b14"
},
"split": "train",
"upstreamId": "0049"
}
},
{
"annotation": {
"bytes": 3394,
"path": "annotations/jawildtext-board-0127.json",
"sha256": "ece060bc790218e7c55af54ca69fffc2699aad5cdb713a3fa149aaed527001c6",
"sourceField": "record.polygons"
},
"category": "board-or-sign",
"evaluation": {
"ignoredTokens": ["<UNK>"],
"mode": "annotation-token-coverage",
"notes": "Polygon order is annotation order, not guaranteed page reading order; score per-polygon text coverage and ignore only literal <UNK> labels."
},
"groundTruth": {
"bytes": 331,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/jawildtext-board-0127.txt",
"sha256": "f4c2f2669a47d98403efb4262b1614ccd8c4c7556fe5f3ae8c5d386edc4259a8",
"sourceField": "record.polygons[*].text",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "jawildtext-board-0127",
"image": {
"bytes": 2501453,
"height": 3000,
"mediaType": "image/jpeg",
"path": "images/jawildtext-board-0127.jpg",
"sha256": "cae8b659138849894bcee4f420e3f54d939c34e4b24b2b6744c480d2439cfe08",
"sourceSha256": "cae8b659138849894bcee4f420e3f54d939c34e4b24b2b6744c480d2439cfe08",
"width": 4000
},
"language": "ja",
"provenance": {
"annotationLicense": "Apache-2.0",
"attribution": "JaWildText by Koki Maeda and Naoaki Okazaki (LLM-jp), released as an ICDAR 2026 benchmark.",
"dataset": "llm-jp/jawildtext",
"licenseEvidenceUrls": [
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/LICENSE",
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The source row excluding image bytes was serialized as canonical UTF-8 JSON; annotation values and polygon coordinates were not changed.",
"Ground truth lists each polygons[*].text value verbatim in source array order, separated by LF, with one terminal LF."
],
"pixelLicense": "Apache-2.0",
"privacyReview": "Public historical-site reconstruction display; no identifiable person or private data is visible.",
"repositoryRevision": "627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceDatasetUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/tree/627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceRecordUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/board_vqa/train-00000-of-00006.parquet",
"sourceShard": {
"bytes": 558557296,
"globalRowIndex": 126,
"path": "board_vqa/train-00000-of-00006.parquet",
"rowGroup": 3,
"rowIndex": 12,
"sha256": "2a3ebcb8fa627d2477b8fd0190b34f4f41bb235c268131a564ccae57e44d4b14"
},
"split": "train",
"upstreamId": "0127"
}
},
{
"annotation": {
"bytes": 520,
"path": "annotations/commons-hagye-station-715.json",
"sha256": "3a4d752b0403c4bbc140842bc7fce343e8caaefa13d43ddf7cdf818cb8be7241",
"sourceField": "manualTranscription"
},
"category": "board-or-sign",
"evaluation": {
"ignoredTokens": [],
"mode": "annotation-token-coverage",
"notes": "The station sign is manually transcribed in visual order; score normalized token coverage across the visible Hangul, Latin, CJK, and Arabic-digit labels."
},
"groundTruth": {
"bytes": 70,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/commons-hagye-station-715.txt",
"sha256": "653a0b2be1543052a42ec83c36c2c8375321a3fc03c48051dd70b4b33624d120",
"sourceField": "manualTranscription",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "commons-hagye-station-715",
"image": {
"bytes": 44677,
"height": 405,
"mediaType": "image/jpeg",
"path": "images/commons-hagye-station-715.jpg",
"sha256": "a9ae819505be17d87393695bdadd1aaff47b0a8b81faec98b38408397942b3dc",
"sourceSha256": "a9ae819505be17d87393695bdadd1aaff47b0a8b81faec98b38408397942b3dc",
"width": 567
},
"language": "ko",
"provenance": {
"annotationLicense": "LicenseRef-Public-Domain",
"attribution": "Hagye01.jpg by Wikimedia Commons user Marcopolis; dedicated to the public domain via PD-user-en.",
"dataset": "Wikimedia Commons",
"licenseEvidenceUrls": [
"https://commons.wikimedia.org/w/index.php?title=File:Hagye01.jpg&oldid=1234274506",
"https://commons.wikimedia.org/w/index.php?title=Template:PD-user-en&oldid=358311026"
],
"modifications": [
"Image copied byte-for-byte from the pinned Wikimedia Commons original; no crop, resize, re-encoding, or metadata rewrite.",
"The visible station-sign text was manually transcribed into canonical UTF-8 JSON without using OCR.",
"Ground truth lists each manually transcribed line in visual order, separated by LF, with one terminal LF."
],
"pixelLicense": "LicenseRef-Public-Domain",
"privacyReview": "Public transit station sign; no person or private data is visible, and embedded EXIF contains no GPS coordinates or owner identity.",
"repositoryRevision": "1234274506",
"sourceDatasetUrl": "https://commons.wikimedia.org/wiki/Category:Train_station_signs_of_Seoul_Subway_Line_7",
"sourceFile": {
"bytes": 44677,
"fileTimestamp": "2009-08-18T13:00:50Z",
"pageId": 7592246,
"sha1": "fd1f0cc88f931af22576c7404270837916c60d8d",
"sha256": "a9ae819505be17d87393695bdadd1aaff47b0a8b81faec98b38408397942b3dc",
"sourceUrl": "https://upload.wikimedia.org/wikipedia/commons/c/cb/Hagye01.jpg"
},
"sourceRecordUrl": "https://commons.wikimedia.org/w/index.php?title=File:Hagye01.jpg&oldid=1234274506",
"upstreamId": "7592246"
}
},
{
"annotation": {
"bytes": 24532,
"path": "annotations/jawildtext-receipt-11120.json",
"sha256": "6a31a58f1a6510c1fcba4a18a212d9b18987e29806890fcfcf86f948572506b8",
"sourceField": "record.polygons"
},
"category": "mobile-receipt",
"evaluation": {
"ignoredTokens": [],
"mode": "annotation-token-coverage",
"notes": "The source supplies polygon transcriptions rather than a page-level reading-order transcript; compare normalized annotated text coverage."
},
"groundTruth": {
"bytes": 813,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/jawildtext-receipt-11120.txt",
"sha256": "119b1f698466d6293fabce54514c970eda70a105c8078e881c06153e84717199",
"sourceField": "record.polygons[*].text",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "jawildtext-receipt-11120",
"image": {
"bytes": 812782,
"height": 1280,
"mediaType": "image/jpeg",
"path": "images/jawildtext-receipt-11120.jpg",
"sha256": "beb8367c33ee91c8b7aa51f9dab491ab20899c55d2d4940756da368df8ac4264",
"sourceSha256": "beb8367c33ee91c8b7aa51f9dab491ab20899c55d2d4940756da368df8ac4264",
"width": 720
},
"language": "ja",
"provenance": {
"annotationLicense": "Apache-2.0",
"attribution": "JaWildText by Koki Maeda and Naoaki Okazaki (LLM-jp), released as an ICDAR 2026 benchmark.",
"dataset": "llm-jp/jawildtext",
"licenseEvidenceUrls": [
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/LICENSE",
"https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The source row excluding image bytes was serialized as canonical UTF-8 JSON; annotation values and polygon coordinates were not changed.",
"Ground truth lists each polygons[*].text value verbatim in source array order, separated by LF, with one terminal LF."
],
"pixelLicense": "Apache-2.0",
"privacyReview": "Cash receipt shows public store details, order lines, and totals; the WAON identifier is masked upstream, and no customer name, signature, or payment-card identifier is visible.",
"repositoryRevision": "627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceDatasetUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/tree/627ca7ea7c224ffe1accff8737991fc2240784fa",
"sourceRecordUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/blob/627ca7ea7c224ffe1accff8737991fc2240784fa/receipt_kie/train-00000-of-00008.parquet",
"sourceShard": {
"bytes": 318241920,
"globalRowIndex": 75,
"path": "receipt_kie/train-00000-of-00008.parquet",
"rowGroup": 2,
"rowIndex": 17,
"sha256": "07c3683d639e327dd4a0d8941990a9a9ce5cc604b9ecb828a760f5586e0f19b4"
},
"split": "train",
"upstreamId": "11120"
}
},
{
"annotation": {
"bytes": 13588,
"path": "annotations/cord-v2-test-0080.json",
"sha256": "c2227948663c9ec73943fe35e485162e4eefbf1a0f17b3e3128e5f4ed804fa6e",
"sourceField": "record.ground_truth.valid_line"
},
"category": "mobile-receipt",
"evaluation": {
"ignoredTokens": [],
"mode": "annotation-token-coverage",
"notes": "CORD provides word boxes grouped into valid_line records; score annotated word coverage because blurred header pixels are intentionally unlabelled."
},
"groundTruth": {
"bytes": 87,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/cord-v2-test-0080.txt",
"sha256": "bf2d38b042e28634c20cb7840d81e671f4e56f0cf490ff330736b0c92fec4808",
"sourceField": "record.ground_truth.valid_line[*].words[*].text",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "cord-v2-test-0080",
"image": {
"bytes": 436299,
"height": 648,
"mediaType": "image/png",
"path": "images/cord-v2-test-0080.png",
"sha256": "e37ed26e77e90f45a0ad33dcb9844c61808ad75e1f4f161debaeb1c45822c3f8",
"sourceSha256": "e37ed26e77e90f45a0ad33dcb9844c61808ad75e1f4f161debaeb1c45822c3f8",
"width": 432
},
"language": "id",
"provenance": {
"annotationLicense": "CC-BY-4.0",
"attribution": "CORD by Seunghyun Park, Seung Shin, Bado Lee, Junyeop Lee, Jaeheung Surh, Minjoon Seo, and Hwalsuk Lee; NAVER Clova.",
"dataset": "naver-clova-ix/cord-v2",
"licenseEvidenceUrls": [
"https://github.com/clovaai/cord/blob/327310ce58c1623255821d062b3a759ff3789e3c/LICENSE-CC-BY",
"https://github.com/clovaai/cord/blob/327310ce58c1623255821d062b3a759ff3789e3c/README.md",
"https://huggingface.co/datasets/naver-clova-ix/cord-v2/blob/7f0115a4b758a71d6473b8d085751692da2fef98/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The original ground_truth JSON string was preserved and also parsed into canonical UTF-8 JSON without changing annotation values.",
"Ground truth lists each valid_line[*].words[*].text value verbatim in source order, separated by LF, with one terminal LF."
],
"pixelLicense": "CC-BY-4.0",
"privacyReview": "Cash receipt contains no customer or payment identifier; identifying header content was already blurred in the upstream test image.",
"repositoryRevision": "7f0115a4b758a71d6473b8d085751692da2fef98",
"sourceDatasetUrl": "https://huggingface.co/datasets/naver-clova-ix/cord-v2/tree/7f0115a4b758a71d6473b8d085751692da2fef98",
"sourceRecordUrl": "https://huggingface.co/datasets/naver-clova-ix/cord-v2/blob/7f0115a4b758a71d6473b8d085751692da2fef98/data/test-00000-of-00001-9c204eb3f4e11791.parquet",
"sourceShard": {
"bytes": 234202795,
"globalRowIndex": 80,
"path": "data/test-00000-of-00001-9c204eb3f4e11791.parquet",
"rowGroup": 0,
"rowIndex": 80,
"sha256": "51c65f1788faff392abe2a0b55b023eb23e9be551c509138eaa3a832514224e7"
},
"split": "test",
"upstreamId": "80"
}
},
{
"annotation": {
"bytes": 2919,
"path": "annotations/clinocr-poor-t7-s2.json",
"sha256": "0154dd0e21735d7e8b1f0c99eb961c452b5900469f710ec969721187a027a8b2",
"sourceField": "record.ground_truth"
},
"category": "photographed-form",
"evaluation": {
"ignoredTokens": [],
"mode": "page-transcript",
"notes": "The upstream ground_truth field is a human-audited page transcript; evaluate grapheme CER and whitespace-normalized word error rate."
},
"groundTruth": {
"bytes": 1203,
"comparisonNormalization": {
"case": "casefold",
"punctuation": "preserve",
"unicode": "NFC",
"whitespace": "collapse-runs-and-trim"
},
"derivation": "verbatim-from-annotation",
"encoding": "UTF-8",
"joinSeparator": "LF",
"path": "ground-truth/clinocr-poor-t7-s2.txt",
"sha256": "df32e31287873c452b4a504085bde178c97e9319fe740b03c616f5a689808139",
"sourceField": "record.ground_truth",
"terminalNewline": true,
"unicodeNormalization": "none",
"whitespaceNormalization": "none"
},
"id": "clinocr-poor-t7-s2",
"image": {
"bytes": 167222,
"height": 1707,
"mediaType": "image/jpeg",
"path": "images/clinocr-poor-t7-s2.jpg",
"sha256": "fd64e12da893cc1f111fade29ba4a826cc3090750d176069f10c1891e53ab4de",
"sourceSha256": "fd64e12da893cc1f111fade29ba4a826cc3090750d176069f10c1891e53ab4de",
"width": 1280
},
"language": "en",
"provenance": {
"annotationLicense": "MIT",
"attribution": "ClinOCR-Bench by Enshuo Hsu, Jin Zhou, and Kirk Roberts; copyright 2026 ClinOCR-Bench.",
"dataset": "ClinOCR-Bench/ClinOCR-Bench",
"licenseEvidenceUrls": [
"https://github.com/ClinOCR-Bench/ClinOCR-Bench/blob/3b720a951bb7eec4a4f4fb34a636e7335a19981e/LICENSE",
"https://github.com/ClinOCR-Bench/ClinOCR-Bench/blob/3b720a951bb7eec4a4f4fb34a636e7335a19981e/README.md",
"https://huggingface.co/datasets/ClinOCR-Bench/ClinOCR-Bench/blob/cb7c0c48a4f3d1c9054fb6548ccef84768983472/README.md"
],
"modifications": [
"Image copied byte-for-byte from the pinned Parquet image.bytes field; no crop, resize, re-encoding, or metadata rewrite.",
"The source row excluding image bytes was serialized as canonical UTF-8 JSON without changing annotation values.",
"The human-audited ground_truth string was copied verbatim and one terminal LF was added to the text file."
],
"pixelLicense": "MIT",
"privacyReview": "Dataset authors identify ClinOCR-Bench as synthetic and protected-health-information-free; displayed names, identifiers, and clinical content are fictional, and no identifiable person is visible.",
"repositoryRevision": "cb7c0c48a4f3d1c9054fb6548ccef84768983472",
"sourceDatasetUrl": "https://huggingface.co/datasets/ClinOCR-Bench/ClinOCR-Bench/tree/cb7c0c48a4f3d1c9054fb6548ccef84768983472",
"sourceRecordUrl": "https://huggingface.co/datasets/ClinOCR-Bench/ClinOCR-Bench/blob/cb7c0c48a4f3d1c9054fb6548ccef84768983472/poor/test-00000-of-00001.parquet",
"sourceShard": {
"bytes": 17486026,
"globalRowIndex": 42,
"path": "poor/test-00000-of-00001.parquet",
"rowGroup": 0,
"rowIndex": 42,
"sha256": "49109997ffd99fcf04f77d1a5507e72ea22543a1be1bc7f35babad572c703f97"
},
"split": "test",
"upstreamId": "poor_t7_s2"
}
}
],
"heldOutPolicy": "These files are regression inputs only and must not be used to tune OCR models or acceptance thresholds after observing their outputs.",
"koreanCohort": {
"fastDisposition": {
"accurateTierResults": {
"balanced": {
"releaseGatePassed": true,
"tokenF1": 0.692308,
"tokenPrecision": 0.9,
"tokenRecall": 0.5625
},
"best": {
"releaseGatePassed": true,
"tokenF1": 0.692308,
"tokenPrecision": 0.9,
"tokenRecall": 0.5625
}
},
"boundedStrategyAudit": {
"diagnosticManifestSha256": "8e82e22b5d939ca40e97d8a74e8ce73dda451e99ba8b33a95a94723c4679d1b4",
"failed": 6,
"tested": 6
},
"decision": {
"enforcedBehavior": "reject-before-tesseract-spawn",
"unsupportedReason": "Fast OCR does not support Korean. Install the Accurate OCR bundle and choose Balanced or Best."
},
"evidence": {
"artifactSha256": "42b9609dab9b8680c208d4b20828314ce912f3691bbd12114b31ab31b4cfbd05",
"fastReportSha256": "0a9743c46e67aad7948d5a5880bfbd5096a4023c6006de5f94668580a2df276a",
"fastText": "=\n< 하 계\n: Hagye 최\nㅜㅠ 좋\n==",
"fastTextSha256": "6ddecdd52dd0a66074fb3c1d003a28493975f63421f70600706ccc01b7bb2c73",
"qualityCheckpointSha256": "41e4c01959602c3c255c09210fd3203ab80eee696bf28578d1137bf07f38bab7",
"sourceImageId": "sha256:127753b7916aea38eac139a4070af95854ccdc68f93a9c4d59618c7e8c7f4bfa",
"verifierSha256": "d4f9e3c1572c7eba4f5157c2a70785233ceed0296cb734a50890e940d16dd2cf"
},
"fastResult": {
"releaseGatePassed": false,
"tokenF1": 0.214286,
"tokenPrecision": 0.25,
"tokenRecall": 0.1875
},
"status": "REJECTED_AFTER_FROZEN_GATE"
},
"fixtureIds": ["commons-hagye-station-715"],
"perImageTierFloors": {
"balanced": {
"minimumTokenF1": 0.56,
"minimumTokenPrecision": 0.65,
"minimumTokenRecall": 0.5
},
"best": {
"minimumTokenF1": 0.6,
"minimumTokenPrecision": 0.68,
"minimumTokenRecall": 0.55
},
"fast": {
"minimumTokenF1": 0.32,
"minimumTokenPrecision": 0.5,
"minimumTokenRecall": 0.25
}
},
"selectionRule": "On 2026-07-13, enumerate original bitmap files in Wikimedia Commons Category:Train station signs of Seoul Subway Line 7, sort by byte size ascending, retain public-domain landscape photographs at least 500×400 with no people or private data and manually legible Hangul, Latin, and Arabic digits, then choose the smallest. Hagye01.jpg is the first eligible file; the smaller public-domain Junggokst01.jpg is only 411×308.",
"selectionStatus": "FROZEN_BEFORE_ANY_OCR_OUTPUT",
"stopPolicy": "If the frozen Korean fixture fails, do not swap or edit the fixture, transcript, or limits; reconsider the Korean Fast model."
},
"licenseFiles": [
{
"bytes": 11357,
"path": "licenses/apache-2.0.txt",
"sha256": "58d1e17ffe5109a7ae296caafcadfdbe6a7d176f0bc4ab01e12a689b0499d8bd",
"sourceUrl": "https://huggingface.co/datasets/llm-jp/jawildtext/resolve/627ca7ea7c224ffe1accff8737991fc2240784fa/LICENSE",
"spdx": "Apache-2.0"
},
{
"bytes": 18650,
"path": "licenses/cc-by-4.0.txt",
"sha256": "7e7170e3cebf88a9f60c7b8421418323c09304da1af4d5e90f4da1dc1c8a2661",
"sourceUrl": "https://raw.githubusercontent.com/clovaai/cord/327310ce58c1623255821d062b3a759ff3789e3c/LICENSE-CC-BY",
"spdx": "CC-BY-4.0"
},
{
"bytes": 1070,
"path": "licenses/mit-clinocr-bench.txt",
"sha256": "0dfbe6a63906c73427ec273521bf182a636ff1cf39e8271697854229dce17b6f",
"sourceUrl": "https://raw.githubusercontent.com/ClinOCR-Bench/ClinOCR-Bench/3b720a951bb7eec4a4f4fb34a636e7335a19981e/LICENSE",
"spdx": "MIT"
}
],
"schemaVersion": 1
}
Reference in New Issue View Git Blame Copy Permalink