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.

For this family org_answer is empty throughout: recovering it means scanning the whole pre-clean family, which is raw family is 61 GB (> 6 GB cap). The cleaned answer is unaffected.

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 — 16 turns on one image

#	question	answer
1	How many chairs are pictured? Give a very brief answer.	2.
2	How many chairs are at the table? Give a very brief answer.	2.
3	How many things are purple? Give a very brief answer.	1.
4	How many curtains are there? Give a very brief answer.	2.
5	How many chairs? Give a very brief answer.	2.
6	How many pillows are on the couch? Give a very brief answer.	1.
7	How many chairs are near the table? Give a very brief answer.	2.
8	How many stuffed animals are visible? Give a very brief answer.	1.
9	How many chairs are visible? Give a very brief answer.	2.
10	How many table legs are on the rug? Give a very brief answer.	1.
11	How many pillows are on the futon? Give a very brief answer.	1.
12	What tall black object is standing in the corner behind the sofa?	A floor lamp.
13	What is sitting on the white pillow?	A brown teddy bear.
14	What is located beneath the window?	A white radiator.
15	What color are the seats of the chairs?	Light blue.
16	What material is the table top made of?	Glass.

Example 2 — 7 turns on one image

#	question	answer
1	How many tow trucks are there? Answer the question using a single word or phrase.	1.
2	How many tow trucks are there?	To determine the number of tow trucks in the image, we need to carefully observe the vehicles present. 1. Identify the vehicles: The image shows a vehicle with the label "POLICE" and "TO SERVE & PROTECT" on its side, along with the number "806" and the emergency number "9-1-1". This vehicle is equipped with a towing apparatus, indicating it is a tow truck. 2. Count the tow trucks: Since only one vehicle in the image is identified as a tow truck, the total number of tow trucks present is one. Therefore, the answer is 1.
3	What large red text is visible on the white sign above the storefront in the background?	metro
4	What number is printed in blue on the door of the tow truck?	806
5	What phrase is written on the side of the tow truck's boom arm?	TO SERVE & PROTECT
6	What emergency number is visible on the lower part of the truck's door?	9-1-1
7	What flag is visible on the door of the tow truck?	Canadian flag

Example 3 — 7 turns on one image

#	question	answer
1	How many wheels are on the plane? Answer the question using a single word or phrase.	3.
2	How many items are in the sky?	2
3	What registration code is written on the underside of the wing?	D-EIEL
4	What text is visible on the fuselage near the tail?	D-EIEL
5	What is the watermark in the bottom right corner?	by C. Bahrens
6	What is visible falling below the airplane?	A small dark object
7	What is the primary color of the airplane?	White

Example 4 — 5 turns on one image

#	question	answer
1	How many dogs are in the picture? Answer the question using a single word or phrase.	1.
2	How many cars are in the picture?	2.
3	What objects are hanging from the rearview mirror?	Two black dice with white dots.
4	What color is the car visible through the passenger window?	Red.
5	What is the color of the car's interior dashboard and door panel?	Grey.

Example 5 — 7 turns on one image

#	question	answer
1	How many teddy bears are visible? Answer the question using a single word or phrase.	6.
2	How many cars have a surfboard on the roof?	0.
3	What kind of stickers surround the bear in the top-left panel?	Red hearts.
4	What kind of stickers surround the bear in the top-right panel?	Colorful flowers.
5	What kind of stickers surround the bear in the bottom-right panel?	Blue and white stars.
6	What visual effect is applied to the bear in the bottom-left panel?	A kaleidoscope or mirrored reflection effect.
7	What accessory is the teddy bear wearing around its neck?	A purple bow tie.

Example 6 — 4 turns on one image

#	question	answer
1	How many people are visible? Answer the question using a single word or phrase.	4.
2	What blue object is flying in the air in the upper left?	A frisbee.
3	What word is printed on the white banner attached to the net?	frisbee
4	What surface are the players standing on?	Sand.

Example 7 — 4 turns on one image

#	question	answer
1	How many zebra are here? Answer the question using a single word or phrase.	0.
2	What is the main object sitting against the wall?	A wooden bench.
3	What decorative pattern is painted on the backrest of the bench?	A floral design featuring green leaves and flowers.
4	What architectural feature is visible in the wall above the bench?	A window with vertical bars.

Example 8 — 4 turns on one image

#	question	answer
1	Question: How many zebras are there? Answer with a number.	0
2	What is the person in the foreground doing?	Performing a skateboard trick in mid-air.
3	What is the person in the background riding?	A bicycle.
4	What color is the skateboarder's t-shirt?	Light orange.

Example 9 — 5 turns on one image

#	question	answer
1	Question: How many people can you see? Answer with a number.	4
2	Question: How many benches are there? Answer with a number.	2
3	What is the man in the foreground wearing?	A red long-sleeved shirt and red shorts.
4	What object is at the man's feet?	A soccer ball.
5	What is the woman sitting on the bench doing?	She is reading a book.

Example 10 — 7 turns on one image

#	question	answer
1	Question: How many people are visible? Answer with a number.	3
2	Question: How many tennis rackets are there? Answer with a number.	1
3	Question: How many chairs can be seen? Answer with a number.	Many
4	What text is visible on the blue banner behind the player?	CHAMPI and BOS
5	What color are the stadium seats in the background?	Red
6	What number is visible on the back of one of the black seats?	1
7	What is the tennis player wearing?	A white shirt and white shorts

Downloads last month: 17

Total file size:

50.4 GB

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

tallyqa_train_cleaned

How this was cleaned

answer vs org_answer