Datasets:
sample_id stringlengths 8 13 | template_name stringclasses 49
values | repository_template stringclasses 49
values | instance_id stringclasses 110
values | image_path stringlengths 28 46 | label_path stringlengths 28 37 |
|---|---|---|---|---|---|
Arabic-1__01 | Arabic-1 | ar_1 | 01 | datasets/ar_1/01/ar_1-01.png | datasets/ar_1/01/answer.json |
Arabic-1__02 | Arabic-1 | ar_1 | 02 | datasets/ar_1/02/ar_1-02.png | datasets/ar_1/02/answer.json |
Arabic-1__03 | Arabic-1 | ar_1 | 03 | datasets/ar_1/03/ar_1-03.png | datasets/ar_1/03/answer.json |
Arabic-1__04 | Arabic-1 | ar_1 | 04 | datasets/ar_1/04/ar_1-04.png | datasets/ar_1/04/answer.json |
Arabic-1__05 | Arabic-1 | ar_1 | 05 | datasets/ar_1/05/ar_1-05.png | datasets/ar_1/05/answer.json |
Arabic-1__06 | Arabic-1 | ar_1 | 06 | datasets/ar_1/06/ar_1-06.png | datasets/ar_1/06/answer.json |
Arabic-1__07 | Arabic-1 | ar_1 | 07 | datasets/ar_1/07/ar_1-07.png | datasets/ar_1/07/answer.json |
Arabic-1__08 | Arabic-1 | ar_1 | 08 | datasets/ar_1/08/ar_1-08.png | datasets/ar_1/08/answer.json |
Arabic-1__09 | Arabic-1 | ar_1 | 09 | datasets/ar_1/09/ar_1-09.png | datasets/ar_1/09/answer.json |
Arabic-1__10 | Arabic-1 | ar_1 | 10 | datasets/ar_1/10/ar_1-10.png | datasets/ar_1/10/answer.json |
Arabic-1__100 | Arabic-1 | ar_1 | 100 | datasets/ar_1/100/ar_1-100.png | datasets/ar_1/100/answer.json |
Arabic-1__11 | Arabic-1 | ar_1 | 11 | datasets/ar_1/11/ar_1-11.png | datasets/ar_1/11/answer.json |
Arabic-1__12 | Arabic-1 | ar_1 | 12 | datasets/ar_1/12/ar_1-12.png | datasets/ar_1/12/answer.json |
Arabic-1__13 | Arabic-1 | ar_1 | 13 | datasets/ar_1/13/ar_1-13.png | datasets/ar_1/13/answer.json |
Arabic-1__14 | Arabic-1 | ar_1 | 14 | datasets/ar_1/14/ar_1-14.png | datasets/ar_1/14/answer.json |
Arabic-1__15 | Arabic-1 | ar_1 | 15 | datasets/ar_1/15/ar_1-15.png | datasets/ar_1/15/answer.json |
Arabic-1__16 | Arabic-1 | ar_1 | 16 | datasets/ar_1/16/ar_1-16.png | datasets/ar_1/16/answer.json |
Arabic-1__17 | Arabic-1 | ar_1 | 17 | datasets/ar_1/17/ar_1-17.png | datasets/ar_1/17/answer.json |
Arabic-1__18 | Arabic-1 | ar_1 | 18 | datasets/ar_1/18/ar_1-18.png | datasets/ar_1/18/answer.json |
Arabic-1__19 | Arabic-1 | ar_1 | 19 | datasets/ar_1/19/ar_1-19.png | datasets/ar_1/19/answer.json |
Arabic-1__20 | Arabic-1 | ar_1 | 20 | datasets/ar_1/20/ar_1-20.png | datasets/ar_1/20/answer.json |
Arabic-1__21 | Arabic-1 | ar_1 | 21 | datasets/ar_1/21/ar_1-21.png | datasets/ar_1/21/answer.json |
Arabic-1__22 | Arabic-1 | ar_1 | 22 | datasets/ar_1/22/ar_1-22.png | datasets/ar_1/22/answer.json |
Arabic-1__23 | Arabic-1 | ar_1 | 23 | datasets/ar_1/23/ar_1-23.png | datasets/ar_1/23/answer.json |
Arabic-1__24 | Arabic-1 | ar_1 | 24 | datasets/ar_1/24/ar_1-24.png | datasets/ar_1/24/answer.json |
Arabic-1__25 | Arabic-1 | ar_1 | 25 | datasets/ar_1/25/ar_1-25.png | datasets/ar_1/25/answer.json |
Arabic-1__26 | Arabic-1 | ar_1 | 26 | datasets/ar_1/26/ar_1-26.png | datasets/ar_1/26/answer.json |
Arabic-1__27 | Arabic-1 | ar_1 | 27 | datasets/ar_1/27/ar_1-27.png | datasets/ar_1/27/answer.json |
Arabic-1__28 | Arabic-1 | ar_1 | 28 | datasets/ar_1/28/ar_1-28.png | datasets/ar_1/28/answer.json |
Arabic-1__29 | Arabic-1 | ar_1 | 29 | datasets/ar_1/29/ar_1-29.png | datasets/ar_1/29/answer.json |
Arabic-1__30 | Arabic-1 | ar_1 | 30 | datasets/ar_1/30/ar_1-30.png | datasets/ar_1/30/answer.json |
Arabic-1__31 | Arabic-1 | ar_1 | 31 | datasets/ar_1/31/ar_1-31.png | datasets/ar_1/31/answer.json |
Arabic-1__32 | Arabic-1 | ar_1 | 32 | datasets/ar_1/32/ar_1-32.png | datasets/ar_1/32/answer.json |
Arabic-1__33 | Arabic-1 | ar_1 | 33 | datasets/ar_1/33/ar_1-33.png | datasets/ar_1/33/answer.json |
Arabic-1__34 | Arabic-1 | ar_1 | 34 | datasets/ar_1/34/ar_1-34.png | datasets/ar_1/34/answer.json |
Arabic-1__35 | Arabic-1 | ar_1 | 35 | datasets/ar_1/35/ar_1-35.png | datasets/ar_1/35/answer.json |
Arabic-1__36 | Arabic-1 | ar_1 | 36 | datasets/ar_1/36/ar_1-36.png | datasets/ar_1/36/answer.json |
Arabic-1__37 | Arabic-1 | ar_1 | 37 | datasets/ar_1/37/ar_1-37.png | datasets/ar_1/37/answer.json |
Arabic-1__38 | Arabic-1 | ar_1 | 38 | datasets/ar_1/38/ar_1-38.png | datasets/ar_1/38/answer.json |
Arabic-1__39 | Arabic-1 | ar_1 | 39 | datasets/ar_1/39/ar_1-39.png | datasets/ar_1/39/answer.json |
Arabic-1__40 | Arabic-1 | ar_1 | 40 | datasets/ar_1/40/ar_1-40.png | datasets/ar_1/40/answer.json |
Arabic-1__41 | Arabic-1 | ar_1 | 41 | datasets/ar_1/41/ar_1-41.png | datasets/ar_1/41/answer.json |
Arabic-1__42 | Arabic-1 | ar_1 | 42 | datasets/ar_1/42/ar_1-42.png | datasets/ar_1/42/answer.json |
Arabic-1__43 | Arabic-1 | ar_1 | 43 | datasets/ar_1/43/ar_1-43.png | datasets/ar_1/43/answer.json |
Arabic-1__44 | Arabic-1 | ar_1 | 44 | datasets/ar_1/44/ar_1-44.png | datasets/ar_1/44/answer.json |
Arabic-1__45 | Arabic-1 | ar_1 | 45 | datasets/ar_1/45/ar_1-45.png | datasets/ar_1/45/answer.json |
Arabic-1__46 | Arabic-1 | ar_1 | 46 | datasets/ar_1/46/ar_1-46.png | datasets/ar_1/46/answer.json |
Arabic-1__47 | Arabic-1 | ar_1 | 47 | datasets/ar_1/47/ar_1-47.png | datasets/ar_1/47/answer.json |
Arabic-1__48 | Arabic-1 | ar_1 | 48 | datasets/ar_1/48/ar_1-48.png | datasets/ar_1/48/answer.json |
Arabic-1__49 | Arabic-1 | ar_1 | 49 | datasets/ar_1/49/ar_1-49.png | datasets/ar_1/49/answer.json |
Arabic-1__50 | Arabic-1 | ar_1 | 50 | datasets/ar_1/50/ar_1-50.png | datasets/ar_1/50/answer.json |
Arabic-1__51 | Arabic-1 | ar_1 | 51 | datasets/ar_1/51/ar_1-51.png | datasets/ar_1/51/answer.json |
Arabic-1__52 | Arabic-1 | ar_1 | 52 | datasets/ar_1/52/ar_1-52.png | datasets/ar_1/52/answer.json |
Arabic-1__53 | Arabic-1 | ar_1 | 53 | datasets/ar_1/53/ar_1-53.png | datasets/ar_1/53/answer.json |
Arabic-1__54 | Arabic-1 | ar_1 | 54 | datasets/ar_1/54/ar_1-54.png | datasets/ar_1/54/answer.json |
Arabic-1__55 | Arabic-1 | ar_1 | 55 | datasets/ar_1/55/ar_1-55.png | datasets/ar_1/55/answer.json |
Arabic-1__56 | Arabic-1 | ar_1 | 56 | datasets/ar_1/56/ar_1-56.png | datasets/ar_1/56/answer.json |
Arabic-1__57 | Arabic-1 | ar_1 | 57 | datasets/ar_1/57/ar_1-57.png | datasets/ar_1/57/answer.json |
Arabic-1__58 | Arabic-1 | ar_1 | 58 | datasets/ar_1/58/ar_1-58.png | datasets/ar_1/58/answer.json |
Arabic-1__59 | Arabic-1 | ar_1 | 59 | datasets/ar_1/59/ar_1-59.png | datasets/ar_1/59/answer.json |
Arabic-1__60 | Arabic-1 | ar_1 | 60 | datasets/ar_1/60/ar_1-60.png | datasets/ar_1/60/answer.json |
Arabic-1__61 | Arabic-1 | ar_1 | 61 | datasets/ar_1/61/ar_1-61.png | datasets/ar_1/61/answer.json |
Arabic-1__62 | Arabic-1 | ar_1 | 62 | datasets/ar_1/62/ar_1-62.png | datasets/ar_1/62/answer.json |
Arabic-1__63 | Arabic-1 | ar_1 | 63 | datasets/ar_1/63/ar_1-63.png | datasets/ar_1/63/answer.json |
Arabic-1__64 | Arabic-1 | ar_1 | 64 | datasets/ar_1/64/ar_1-64.png | datasets/ar_1/64/answer.json |
Arabic-1__65 | Arabic-1 | ar_1 | 65 | datasets/ar_1/65/ar_1-65.png | datasets/ar_1/65/answer.json |
Arabic-1__66 | Arabic-1 | ar_1 | 66 | datasets/ar_1/66/ar_1-66.png | datasets/ar_1/66/answer.json |
Arabic-1__67 | Arabic-1 | ar_1 | 67 | datasets/ar_1/67/ar_1-67.png | datasets/ar_1/67/answer.json |
Arabic-1__68 | Arabic-1 | ar_1 | 68 | datasets/ar_1/68/ar_1-68.png | datasets/ar_1/68/answer.json |
Arabic-1__69 | Arabic-1 | ar_1 | 69 | datasets/ar_1/69/ar_1-69.png | datasets/ar_1/69/answer.json |
Arabic-1__70 | Arabic-1 | ar_1 | 70 | datasets/ar_1/70/ar_1-70.png | datasets/ar_1/70/answer.json |
Arabic-1__71 | Arabic-1 | ar_1 | 71 | datasets/ar_1/71/ar_1-71.png | datasets/ar_1/71/answer.json |
Arabic-1__72 | Arabic-1 | ar_1 | 72 | datasets/ar_1/72/ar_1-72.png | datasets/ar_1/72/answer.json |
Arabic-1__73 | Arabic-1 | ar_1 | 73 | datasets/ar_1/73/ar_1-73.png | datasets/ar_1/73/answer.json |
Arabic-1__74 | Arabic-1 | ar_1 | 74 | datasets/ar_1/74/ar_1-74.png | datasets/ar_1/74/answer.json |
Arabic-1__75 | Arabic-1 | ar_1 | 75 | datasets/ar_1/75/ar_1-75.png | datasets/ar_1/75/answer.json |
Arabic-1__76 | Arabic-1 | ar_1 | 76 | datasets/ar_1/76/ar_1-76.png | datasets/ar_1/76/answer.json |
Arabic-1__77 | Arabic-1 | ar_1 | 77 | datasets/ar_1/77/ar_1-77.png | datasets/ar_1/77/answer.json |
Arabic-1__78 | Arabic-1 | ar_1 | 78 | datasets/ar_1/78/ar_1-78.png | datasets/ar_1/78/answer.json |
Arabic-1__79 | Arabic-1 | ar_1 | 79 | datasets/ar_1/79/ar_1-79.png | datasets/ar_1/79/answer.json |
Arabic-1__80 | Arabic-1 | ar_1 | 80 | datasets/ar_1/80/ar_1-80.png | datasets/ar_1/80/answer.json |
Arabic-1__81 | Arabic-1 | ar_1 | 81 | datasets/ar_1/81/ar_1-81.png | datasets/ar_1/81/answer.json |
Arabic-1__82 | Arabic-1 | ar_1 | 82 | datasets/ar_1/82/ar_1-82.png | datasets/ar_1/82/answer.json |
Arabic-1__83 | Arabic-1 | ar_1 | 83 | datasets/ar_1/83/ar_1-83.png | datasets/ar_1/83/answer.json |
Arabic-1__84 | Arabic-1 | ar_1 | 84 | datasets/ar_1/84/ar_1-84.png | datasets/ar_1/84/answer.json |
Arabic-1__85 | Arabic-1 | ar_1 | 85 | datasets/ar_1/85/ar_1-85.png | datasets/ar_1/85/answer.json |
Arabic-1__86 | Arabic-1 | ar_1 | 86 | datasets/ar_1/86/ar_1-86.png | datasets/ar_1/86/answer.json |
Arabic-1__87 | Arabic-1 | ar_1 | 87 | datasets/ar_1/87/ar_1-87.png | datasets/ar_1/87/answer.json |
Arabic-1__88 | Arabic-1 | ar_1 | 88 | datasets/ar_1/88/ar_1-88.png | datasets/ar_1/88/answer.json |
Arabic-1__89 | Arabic-1 | ar_1 | 89 | datasets/ar_1/89/ar_1-89.png | datasets/ar_1/89/answer.json |
Arabic-1__90 | Arabic-1 | ar_1 | 90 | datasets/ar_1/90/ar_1-90.png | datasets/ar_1/90/answer.json |
Arabic-1__91 | Arabic-1 | ar_1 | 91 | datasets/ar_1/91/ar_1-91.png | datasets/ar_1/91/answer.json |
Arabic-1__92 | Arabic-1 | ar_1 | 92 | datasets/ar_1/92/ar_1-92.png | datasets/ar_1/92/answer.json |
Arabic-1__93 | Arabic-1 | ar_1 | 93 | datasets/ar_1/93/ar_1-93.png | datasets/ar_1/93/answer.json |
Arabic-1__94 | Arabic-1 | ar_1 | 94 | datasets/ar_1/94/ar_1-94.png | datasets/ar_1/94/answer.json |
Arabic-1__95 | Arabic-1 | ar_1 | 95 | datasets/ar_1/95/ar_1-95.png | datasets/ar_1/95/answer.json |
Arabic-1__96 | Arabic-1 | ar_1 | 96 | datasets/ar_1/96/ar_1-96.png | datasets/ar_1/96/answer.json |
Arabic-1__97 | Arabic-1 | ar_1 | 97 | datasets/ar_1/97/ar_1-97.png | datasets/ar_1/97/answer.json |
Arabic-1__98 | Arabic-1 | ar_1 | 98 | datasets/ar_1/98/ar_1-98.png | datasets/ar_1/98/answer.json |
Arabic-1__99 | Arabic-1 | ar_1 | 99 | datasets/ar_1/99/ar_1-99.png | datasets/ar_1/99/answer.json |
FormStruct-Bench
Dataset Description
FormStruct-Bench is a multilingual benchmark for extracting the semantic and spatial structure of forms from document images. The repository combines a 7,000-page main benchmark, a controlled visual-degradation set, and template-level layout annotations. It supports evaluation of vision-language models and document AI systems on hierarchical key-value extraction, document structure recovery, region localization, table and line-item understanding, and selection-widget interpretation.
The main task takes a single form image as input and predicts its complete hierarchical answer tree. Template annotations provide complementary pixel-space regions and layout metadata for structure-aware evaluation.
Rights-audit notice (2026-07-23): Apache-2.0 applies to the associated software repository only, not to this dataset. The source, rights, and privacy evidence needed to authorize redistribution has not yet been verified for every template. Consult
DATA_LICENSE.mdandprovenance/template_rights.csvbefore accessing or redistributing files. Rows markedUNVERIFIED_DO_NOT_REDISTRIBUTEare not cleared for redistribution. Some source material is reported under CC BY-NC-SA 4.0 or other terms; those terms apply only where the per-template record links verifiable evidence.
Repository Contents
| Component | Scope | Description |
|---|---|---|
datasets/ |
70 canonical templates + 10 redundant directories | 7,000 official benchmark pages; redundant directories are excluded |
dataset-augment/ |
1,216 degraded pages | Controlled visual robustness data |
template_annotation/ |
70 benchmark templates + 10 redundant templates | Template-level fields, boxes, and layout metadata |
splits/template_stratified_seed42/ |
70 templates, 7,000 indexed pages | Official train/validation/test assignments |
provenance/template_rights.csv |
80 template records | Per-template source, rights, privacy, and redistribution audit status |
The repository currently retains directories for 80 templates, but the main
benchmark contains 100 filled instances for each of 70 canonical templates
(7,000 pages). The 10 additional directories correspond to the redundant
templates listed below and are excluded from the official benchmark and all
splits. The augmentation data contains
degraded variants of selected source pages. The template annotations contain
annotations for all 70 main-benchmark templates plus 10 redundant templates.
The 70 templates in datasets/ define the canonical dataset scope. The extra
annotation files are retained only as redundant data and are not part of the
formal benchmark.
Main Benchmark
Statistics
| Property | Value |
|---|---|
| Templates | 70 |
| Instances per template | 100 |
| Total document pages | 7,000 |
Valid PNG and answer.json pairs |
7,000 |
| Total leaf fields | 256,806 |
| Empty leaf fields | 0 |
| Unique canonical answers | 6,921 |
| Unique image pixel hashes | 6,946 |
Each main sample contains the same answer in three representations:
answer.json: machine-readable hierarchical key-value data;answer.md: a human-readable nested list; andanswer.html: a browser-renderable hierarchical view.
Language Distribution
| Language | Script | Direction | Templates | Instances |
|---|---|---|---|---|
| Japanese | Han, Hiragana, Katakana | LTR | 22 | 2,200 |
| English | Latin | LTR | 19 | 1,900 |
| Chinese | Han | LTR | 11 | 1,100 |
| Arabic | Arabic | RTL | 8 | 800 |
| Spanish | Latin | LTR | 3 | 300 |
| Portuguese | Latin | LTR | 3 | 300 |
| German | Latin | LTR | 2 | 200 |
| Chinese-English | Han and Latin | LTR | 2 | 200 |
| Total | 70 | 7,000 |
The filename prefix zn is retained from the original data and denotes
Chinese; it is not an ISO 639 language code. zn_en denotes bilingual
Chinese-English templates.
Directory Structure
datasets/
{template_name}/
{instance_id}/
{template_name}-{instance_id}.png
answer.json
answer.md
answer.html
Example:
datasets/en_1/01/
en_1-01.png
answer.json
answer.md
answer.html
Answer Format
answer.json stores the semantic answer as a nested JSON object. Internal
objects represent sections or semantic groups, while leaf values contain the
text associated with individual form fields.
{
"PRODUCT SPECIFICATION": {
"Brand": "Marlboro",
"Company": "Philip Morris International",
"Country": "United States"
},
"Prepared by": "Laura Bennett",
"Date": "28/02/2024"
}
The schema varies by template and can also vary across instances of the same template. Systems should therefore predict the full hierarchy instead of assuming one fixed global field schema.
Template Annotations
template_annotation/ contains 80 standalone JSON files with one reviewed
template annotation per file. Exactly 70 files correspond to the canonical
templates in the main dataset. The following 10 files are redundant data and
must be excluded from official dataset statistics, splits, training scope, and
evaluation:
de_3.json
de_4.json
es_4.json
ja_23.json
ja_24.json
ja_25.json
ja_26.json
ja_27.json
ja_28.json
zn_12.json
The annotation directory as a whole covers Arabic, German, English, Spanish,
Japanese, Portuguese, Chinese, and bilingual Chinese-English forms. Unless a
separate exploratory use explicitly requires the redundant files, consumers
should join annotations against the 70 template names present in datasets/.
Annotation Statistics
The structural statistics below use only the 70 canonical benchmark annotations and exclude the 10 redundant files.
| Property | Value |
|---|---|
| Annotation files | 80 |
| Canonical benchmark annotations | 70 |
| Redundant annotations | 10 |
| Regions per canonical template | 1-9 (mean 4.59) |
| Fields per canonical template | 14-86 (mean 42.69) |
| Local grids per canonical template | 0-2 (mean 0.20) |
Canonical portrait templates (864 x 1232) |
63 |
Canonical landscape templates (1232 x 864) |
7 |
Annotation Schema
Each file contains:
id: template identifier;img: source-image reference from the annotation environment;original_width,original_height: page dimensions in pixels;fields: recursive field annotations;semantic_key: normalized semantic field name;original_label: label in the source document language;bbox: pixel-space box in[x_min, y_min, x_max, y_max]format;data_type: types such astext,number,checkbox, andcheckbox_multi;valueorvalues: one or more associated value or option regions;keys: nested child fields; andmetadata: structural, visual, domain, language, difficulty, section, region, table-region, and line-item-group metadata.
Shortened example:
{
"id": 182,
"img": "/data/upload/2/49a188a5-en_1.jpg",
"original_width": 864,
"original_height": 1232,
"fields": [
{
"semantic_key": "Brand",
"original_label": "Brand",
"bbox": [84, 192, 129, 208],
"data_type": "text",
"value": {
"bbox": [142, 190, 335, 207],
"data_type": "text"
}
}
],
"metadata": {
"language": "English",
"domain": "business",
"layout_structure": {
"page_bbox": [0, 0, 864, 1232]
}
}
}
The img entries are internal annotation-system paths, not downloadable URLs.
Use the JSON filenames to associate template annotations with matching template
names in the main benchmark.
Visual-Degradation Data
dataset-augment/ supports controlled robustness evaluation. It contains 76
source pages with:
- 1,140 factorial variants from five degradation families at three severity levels; and
- 76 additional
combineddegradation images.
The five factorial degradation families are:
| Variant | Effect |
|---|---|
blur_noise |
Blur, image noise, salt-and-pepper noise, motion blur, and JPEG artifacts |
dilate |
Thickened foreground ink or table lines with controlled local bending |
erode |
Thinned or faded foreground ink and table lines |
perspective_skew |
Rotation, translation, scale, and perspective displacement |
occlusion_stain |
Stains, shadows, creases, and partial occlusion |
Factorial variants use low, medium, and high severity levels. Their
directory structure is:
dataset-augment/
{template_name}/
{source_instance_id}/
{variant}/
{level}/
{template_name}-{source_instance_id}.png
answer.json
augment_meta.json
Each augment_meta.json records the deterministic seed, transformation
parameters, source and output sizes, before/after image metrics, and pixel
difference statistics. The 76 top-level combined images have image and
augmentation metadata but do not include an answer.json sidecar. Evaluation
code should pair only samples that have the required clean source and answer.
Loading the Data
The repository uses a task-specific directory structure rather than a single tabular file. A minimal Python loader for the main benchmark is:
import json
from pathlib import Path
root = Path("datasets")
samples = []
for image_path in sorted(root.glob("*/*/*.png")):
answer_path = image_path.parent / "answer.json"
if not answer_path.is_file():
continue
samples.append(
{
"template": image_path.parent.parent.name,
"instance_id": image_path.parent.name,
"image_path": str(image_path),
"answer": json.loads(answer_path.read_text(encoding="utf-8")),
}
)
print(len(samples)) # 7000
Template annotations can be loaded independently:
annotation_root = Path("template_annotation")
annotations = {
path.stem: json.loads(path.read_text(encoding="utf-8"))
for path in sorted(annotation_root.glob("*.json"))
}
Tasks and Evaluation
The repository is suitable for:
- image-to-hierarchical-JSON extraction;
- form key-value extraction with full semantic paths;
- document schema and hierarchy recovery;
- region and line-item-group localization;
- table, widget, and key-value relation analysis;
- multilingual and right-to-left form understanding; and
- robustness evaluation under controlled visual degradation.
Relevant evaluation families include whole-page exact match, normalized schema tree-edit similarity, normalized value edit similarity, path-sensitive field accuracy, region F1 at an IoU threshold, line-item-group F1, and widget answer accuracy. Evaluation code and exact metric definitions are maintained in the associated FormStruct-Bench project repository.
Splits
FormStruct-Bench defines one official, fixed, template-disjoint split generated with seed 42. The split used in the paper is:
| Split | Templates | Pages | Human-review status |
|---|---|---|---|
| Train | 49 | 4,900 | Not claimed as fully reviewed |
| Validation | 10 | 1,000 | Not claimed as fully reviewed |
| Test | 11 | 1,100 | Fully reviewed |
The authoritative release files are:
splits/template_stratified_seed42/train_index.jsonl;splits/template_stratified_seed42/val_index.jsonl;splits/template_stratified_seed42/test_index.jsonl.
The JSONL paths use the public repository's normalized template directory names and resolve from the repository root. The 10 redundant templates are excluded from every split. Do not randomly split pages: instances from the same template share substantial visual and semantic structure and would leak across partitions. Results should report the dataset revision and use these manifests.
Data Creation and Provenance
The repository contains populated form pages, hierarchical answer sidecars, reviewed template annotations, and deterministic visual augmentations. The augmentation metadata records seeds and parameters for reproducibility.
provenance/template_rights.csv is the controlling, per-template audit record
for source title and URL, rightsholder, source license and evidence, privacy
review, and redistribution status. It covers all 70 canonical templates and
the 10 redundant templates. Blank evidence fields mean that the fact has not
been established; they do not mean public domain or unrestricted use.
At the 2026-07-23 release audit, the repository metadata did not contain enough
machine-readable evidence to close these fields for any template. Those rows
are conservatively marked UNVERIFIED_DO_NOT_REDISTRIBUTE. The paper reports
that some materials are subject to CC BY-NC-SA 4.0 or mixed copyright terms,
but an aggregate statement cannot establish which terms govern an individual
file. A row may be changed to CLEARED only after it names the source,
rightsholder, applicable license, evidence URL or archived permission, and a
completed privacy review. See provenance/README.md for the audit procedure.
Data Quality and Limitations
- The benchmark contains many pages per template but only 70 main templates. Page-level results are therefore clustered and should not be interpreted as 7,000 independent document designs.
- The language and domain distributions are imbalanced. Japanese and English account for most main templates.
- Exact pixel hashing finds 6,946 unique images among 7,000 pages. Duplicate or near-duplicate images can affect evaluation if splits are created without template and duplicate controls.
- Some source
answer.jsonfiles contain repeated object keys. Standard JSON parsers retain only the last repeated key, while the Markdown and HTML sidecars may preserve all repeated entries. Consumers should audit this behavior for their task. - The 1,216 augmented images form a selected robustness collection, not a degradation of all 7,000 main pages. Downstream evaluation may use a smaller subset after clean-pair validation.
- The 10 redundant annotation files listed above are outside the canonical 70-template scope. They must not be included in official benchmark results; join components by template name rather than assuming identical coverage.
- The repository currently also retains
datasets/directories for those 10 redundant templates. Their presence is archival only: they are excluded from the official 7,000-page count and every released split. - Template annotation
imgvalues are non-portable internal paths. - Automatically derived difficulty and visual metadata should not be treated as independently validated human judgments.
- Form fields and populated values may resemble personal, financial, medical, employment, or government information. The data must not be used to make decisions about real individuals.
Privacy and Responsible Use
The forms include identity-like names and values as well as fields associated
with potentially sensitive domains. The current release records do not
establish that every value is synthetic or anonymized. This is tracked per
template in provenance/template_rights.csv; an unverified row is not
privacy-cleared. Report suspected personal or sensitive information through
the repository's Community tab and identify the template and instance so
maintainers can remove or quarantine it.
FormStruct-Bench is intended for document AI research and system evaluation. It is not intended for identity verification, eligibility decisions, surveillance, or automated decisions that affect individuals.
License and Rights
This dataset has no blanket Apache-2.0 license. Hugging Face metadata uses
license: other because rights are mixed and source-specific:
- Apache-2.0 covers only code in the associated software repository.
- CC BY-NC-SA 4.0 applies only to files whose per-template audit row cites evidence for that license; its attribution, non-commercial, and share-alike conditions remain in force.
- Base images and document designs under other terms remain subject to those source terms.
- Answers and annotations may be derivative of the underlying form, and augmented images inherit restrictions from their clean source image.
UNVERIFIED_DO_NOT_REDISTRIBUTEmeans that the release does not provide a verified grant of redistribution rights for that row.
DATA_LICENSE.md defines the component-level policy and
provenance/template_rights.csv is the controlling per-template record. Access
to repository files does not itself grant copyright, privacy, publicity,
trademark, database, or other rights.
Citation
No canonical citation is included in the current repository. When reporting results, cite the Hugging Face dataset repository and record the exact commit or revision used. Add the associated paper's BibTeX entry here when it becomes available.
Maintenance
Questions, corrections, and data-quality reports should be submitted through the Hugging Face dataset repository's Community tab. Versioned releases should document changes to template coverage, answer files, annotations, and augmentation metadata.
- Downloads last month
- 4,288