datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
funsdfunsd-layoutlmv3funsd_plus
Dataset Card for "funsd_plus"
Additional Information
The FUNSD dataset provides ground truth data for form understanding.
Following research, the FUNSD+ dataset addresses inconsistencies in labeling found in the original FUNSD Vu and Nguyen (2020), which could hinder its effectiveness for key-value extraction.
FUNSD+ significantly expands the document count from 199 to 1,113, while enhancing header, question, and answer statistics as outlined in the provided table.… See the full description on the dataset page: https://huggingface.co/datasets/konfuzio/funsd_plus.funsd-json
Dataset Card for FUNSD (JSON Format)
This dataset contains the preprocessed JSON format of the FUNSD dataset, designed for form understanding in noisy scanned documents. It includes the original structure with annotations in JSON format, as per the original FUNSD dataset. This dataset is intended for document understanding tasks such as OCR, layout parsing, and key-value extraction.
Dataset Details
Dataset Description
The FUNSD (Form Understanding… See the full description on the dataset page: https://huggingface.co/datasets/Tamilnilavan/funsd-json.funsdfunsd-image-featureFUNSDexp135funsd-json
Dataset Card for FUNSD (JSON Format)
This dataset contains the preprocessed JSON format of the FUNSD dataset, designed for form understanding in noisy scanned documents. It includes the original structure with annotations in JSON format, as per the original FUNSD dataset. This dataset is intended for document understanding tasks such as OCR, layout parsing, and key-value extraction.
Dataset Details
Dataset Description
The FUNSD (Form Understanding in… See the full description on the dataset page: https://huggingface.co/datasets/davidle7/funsd-json.FUNSD-vu2020revising
Dataset Card for FUNSD-vu2020revising
Dataset Summary
This is the revised version of the FUNSD dataset as proposed by Vu, H. M., & Nguyen, D. T. N. (2020).
Supported Tasks and Leaderboards
The Form Understanding challenge comprises three tasks, namely word grouping, semantic-entity labeling, and entity linking.
Dataset Structure
Data Instances
More Information Needed
Data Fields
The data fields are the… See the full description on the dataset page: https://huggingface.co/datasets/dhdeco/FUNSD-vu2020revising.FUNSD_rejected
FUNSD — FUNSD_rejected
Rejection-sampled from the FUNSD train split. This split holds the rejected items — the answer field holds the official ground truth.
rows
115
QA pairs
460
shards
1
accepted / rejected (whole family)
911 / 460
accept rate
66.4%
verifier
exact
The rejected split is training data, not just diagnostics: answer is the official ground truth, and wrong_vlm records what the model said instead.
How the data was produced… See the full description on the dataset page: https://huggingface.co/datasets/elliot-mllm/FUNSD_rejected.usa_passports_ocr_funsdfunsd_train_cleaned
funsd_train_cleaned
The funsd_train family of the ElliotVL supervised-fine-tuning pool, after VLM cleaning.
images
363
QA turns
6,820
answers rewritten by the cleaning pass
33
QA created by the cleaning pass (new_qa)
3,750 (55.0%)
shards
1
How this was cleaned
A vision-language model read each image together with its QA and judged the item. The pass is
not a filter that only removes rows — it rewrites answers it finds wrong but… See the full description on the dataset page: https://huggingface.co/datasets/Elliot-Data/funsd_train_cleaned.FUNSD_RS_nothink
FUNSD — FUNSD_RS_nothink
Rejection-sampled from the FUNSD train split. This split holds the accepted items, answer only.
rows
125
QA pairs
911
shards
1
accepted / rejected (whole family)
911 / 460
accept rate
66.4%
verifier
exact
How the data was produced
A VLM answers every question at temperature 0 with reasoning enabled. Its answer is compared with
the official ground truth by the verifier described below; matches go to… See the full description on the dataset page: https://huggingface.co/datasets/elliot-mllm/FUNSD_RS_nothink.FUNSD_RS_think
FUNSD — FUNSD_RS_think
Rejection-sampled from the FUNSD train split. This split holds the accepted items, with the model's reasoning trace.
rows
125
QA pairs
911
shards
1
accepted / rejected (whole family)
911 / 460
accept rate
66.4%
verifier
exact
How the data was produced
A VLM answers every question at temperature 0 with reasoning enabled. Its answer is compared with
the official ground truth by the verifier described below; matches… See the full description on the dataset page: https://huggingface.co/datasets/elliot-mllm/FUNSD_RS_think.funsd
Dataset Card for "funsd"
More Information needed
funsd_donut
Dataset Card for "funsd_donut"
More Information needed
funsd-layoutlmv3
Neural Metrics · Noisy scanned forms with key-value ground truth.
FUNSD is a small, deliberately grubby set of scanned forms annotated with entities and the links between them. It is the classic sanity check for form understanding.
We use it for: key-value pair extraction - entity linking on forms - checking that a model degrades gracefully on genuinely bad scans.
Attribution
This is an unmodified fork of nielsr/funsd-layoutlmv3, created by the Qwen team.… See the full description on the dataset page: https://huggingface.co/datasets/NeuralMetrics/funsd-layoutlmv3.FUNSDExp50donut-funsd
Dataset Card for "donut-funsd"
More Information needed
funsdfunsd-layoutlmv3FUNSD_datasetfunsd-bioes
Dataset Card for "funsd-bies"
More Information needed
FUNSD
Dataset Card for "FUNSD"
More Information needed
FUNSDFUNSD_longformfunsd-bank-paragraph-test3funsd-bank-paragraph-test3-encodedfunsd-bank-paragraph-test4-encodedfunsd