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 salvageable, drops what it cannot salvage, and adds QA where the image supports more than the source provided. Each row carries the judge's own record in clean_meta, including the cleaner identity, the policy it applied, and its per-item scores for legibility, richness and coverage.

A large share of the QA here was written by the cleaning pass, not by the original dataset. Across the pool that share runs from roughly half to over 80% of a family's turns, and it is reported in the table above. Those rows have an empty org_answer because no pre-clean original exists. Treat them as model-generated supervision: they were judged against the image, but they are not human annotation, and model-written QA is where formatting defects are most likely.

The effect on the answers that were carried over is substantive rather than cosmetic. In this pool the pass corrects values, not just wording — a curve's critical point restated from 4.00 to 2.00, a computed ratio from 1 to Approximately 1.33 — and for some families it removed the content entirely, which is why those families are absent here rather than published empty.

`answer` vs `org_answer`

answer — the cleaned answer. Train on this.
org_answer — the pre-cleaning answer from the same (image, question) in the uncleaned pool. It is empty for QA the cleaning pass added, which have no original.

Where the two differ, the difference is the correction. Keeping both makes every edit auditable instead of asking you to take the cleaning on trust.

This dataset supersedes an earlier upload of the same family that was built from the pre-clean pool. That version carried the uncorrected answers and has been deleted.

Schema

image — HF Image(), renders directly in the dataset viewer
image_sha256 — content hash of the image
n_qa — number of turns attached to this image
qa — list of {question, answer, org_answer}, in source order
family / clean_meta — provenance and the judge's per-row record

A row with more than one entry in qa is a conversation over a single image, not a set of independent questions — the turns depend on each other and the image belongs to all of them. Keep them together and in order when training.

Parquet row groups are written at ~32 MB so the dataset viewer pages without stalling.

What is NOT claimed

No rejection sampling was run against this family: there is no accept/reject verdict per row, and answer is the cleaned reference rather than a model generation that a string verifier confirmed. For that, use the *_RS_think / *_rejected datasets in this org.

Examples

10 rows taken straight from this dataset. Images are the original files as stored here - no downscaling, no recompression - and every turn is shown in full.

Example 1 — 4 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	Footpath To Colchester and Greenstead
2	Which direction does the arrow on the sign point?	Left
3	What are the colors of the sign's background and border?	White background with a black border
4	How is the sign supported?	It is mounted on two white posts

Example 2 — 3 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	DELL
2	What is the shape and color of the badge containing the text?	A silver or metallic circle.
3	Describe the texture of the dark surface surrounding the logo.	It has a grainy, speckled texture.

Example 3 — 3 turns on one image

#	question	answer
1	Describe the logo centered above the main text.	A red letter 'F' inside a blue circle, enclosed within a yellow oval.
2	What objects are arranged in rows directly below the main blue sign?	Beer cans featuring a black and white checkered pattern.
3	What text is visible on the strip at the very bottom of the image?	FOSTER'S

Example 4 — 5 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	STIRLING CASTLE ARGYLL'S LODGINGS
2	What directions do the arrows on the sign point?	The top arrow points left, and the bottom arrow points right.
3	What is the sign mounted on?	A rough stone wall.
4	What white symbols appear on the sign besides the text and arrows?	Icons of walking people (pedestrians).
5	Is there any visible damage on the sign?	Yes, there is a patch of peeling paint or damage on the word 'LODGINGS'.

Example 5 — 5 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	GOOD HOME MADE FOOD CASK CONDITIONED ALES GARDEN
2	What are the colors of the text and the background of the sign?	White text on a dark background.
3	What graphical elements separate the different sections of text?	Horizontal white lines.
4	What is visible in the reflection on the upper part of the sign?	A building or houses.
5	What color is the border surrounding the sign?	Gold or brass.

Example 6 — 3 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	SCOTTISH NATIONAL GALLERY MODERN
2	What large architectural structures are visible behind the banners?	Large, fluted stone columns.
3	What is visible in the background on the left side of the image?	A person standing near a window and a dark sculpture.

Example 7 — 4 turns on one image

#	question	answer
1	Transcribe all legible text in this image. Use one line per text instance.	Information Tickets Trains
2	What symbols appear to the right of the text on each panel?	Blue arrows pointing diagonally up and to the right.
3	What surface is the sign mounted on?	A dark brick wall.
4	How many horizontal sections does the sign consist of?	Three.

Example 8 — 4 turns on one image

#	question	answer
1	What symbol is located at the top center of the sign?	A black triangle containing an exclamation mark.
2	What is the sign attached to?	A wire mesh fence.
3	What is the background color of the sign?	Yellow.
4	What small text is visible in the bottom left corner of the sign?	E 1603 6182 1

Example 9 — 4 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Transcribe all legible text in this image. Use one line per text instance.	Marlboro LIGHTS	arlboro LIGHTS
2	Describe the layout and font style of the text in the image.	The text is arranged in two lines: 'Marlboro' is on top in a large serif font, and 'LIGHTS' is centered below it in a smaller, all-caps serif font.	(unchanged)
3	What visual artifact affects the left side of the image?	There is a bright glare or reflection on the left side that partially obscures the letters 'M' and 'a' in 'Marlboro'.	(unchanged)
4	What are the colors of the text and the background?	The text is black and the background is white.	(unchanged)

Example 10 — 3 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Transcribe all legible text in this image. Use one line per text instance.	SyncMaster 213T	SyncMaster 213T
2	What is the color of the monitor's bezel?	Silver	(unchanged)
3	What interface element is visible on the right edge of the screen?	A scroll bar with up and down arrows	(unchanged)

Downloads last month: 21

Total file size:

153 MB

You need to agree to share your contact information to access this dataset

icdar2013_train_cleaned

How this was cleaned

answer vs org_answer