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 17 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 — 6 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'a monitor device on table.' information within this image?	[0.58, 0.58, 0.67, 0.93]
2	What can you infer from the mentioned visual clues? Please respond briefly.	There is a baby being monitored somewhere else in house.
3	What animal is the man on the right holding?	A small, scruffy dog.
4	What are the people on the left looking at?	An open photo album or book.
5	What beverage container is visible on the table near the man with the beard?	A brown glass beer bottle.
6	What lighting sources are visible on the table?	Several candles, including a large white pillar candle and smaller tealights.

Example 2 — 6 turns on one image

#	question	answer
1	Please identify the area in this image where 'a group of people gathered on a dirt field with one of them holding a soccer ball.' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.76, 0.00, 0.89, 0.73], [0.76, 0.17, 0.81, 0.29]
2	What can you infer from the mentioned visual clues? Please respond briefly.	they were playing a soccer match.
3	Please identify the area in this image where 'someone in camo pants holding a rifle while looking at an unarmed person.' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.05, 0.07, 0.17, 0.58], [0.07, 0.17, 0.17, 0.42]
4	What is the man in the foreground wearing and what is he carrying?	He is wearing a light green shirt and blue jeans, and he is carrying a cardboard box.
5	What text is visible in the bottom left corner of the image?	FANDANGO MOVIECLIPS
6	Describe the man standing in the center of the group.	He is shirtless and wearing camouflage pants.

Example 3 — 7 turns on one image

#	question	answer
1	I need assistance finding the 'trash all over a desk' in this image. Could you provide me with the corresponding region coordinates [xmin, ymin, xmax, ymax]?	[0.58, 0.47, 0.75, 0.73]
2	What can you infer from the mentioned visual clues? Please respond briefly.	the owner of the desk has had a busy day and didn't have time to clean
3	Please identify the area in this image where 'a woman reaching out to shake a man's hand' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.13, 0.08, 0.62, 0.99]
4	What text is visible in the bottom left corner of the image?	MOVIECLIPS.COM
5	What animals are depicted in the framed picture hanging on the wall?	Camels
6	What items are hanging on the coat rack to the left of the door?	A hat, a scarf, and a jacket
7	What object is mounted on the wall above the door?	A clock

Example 4 — 5 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'items on table between two man seated.' information within this image?	[0.51, 0.66, 0.63, 1.00]
2	What can you infer from the mentioned visual clues? Please respond briefly.	they are having a conversation or meeting at a table.
3	Describe the visible injuries on the man facing the camera.	He has a bruised/swollen eye (black eye) and a cut on his lip.
4	What is visible about the person seated on the left side of the image?	Only their back and side profile are visible; they are wearing a grey jacket and glasses.
5	What can be seen through the large window in the background?	A street scene with a car, a building facade, and people standing outside.

Example 5 — 7 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'booth seating' information within this image?	[0.00, 0.61, 1.00, 0.89]
2	What can you infer from the mentioned visual clues? Please respond briefly.	this is a sit down restaurant
3	Please identify the area in this image where 'salt shaker' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.65, 0.65, 0.70, 0.83]
4	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'drink glasses' information within this image?	[0.01, 0.56, 0.05, 0.68]
5	What text is visible as a watermark in the bottom left corner of the image?	MOVIECLIPS.COM
6	Describe the jacket worn by the man sitting in the booth.	He is wearing a track jacket with grey, white, and navy blue color blocks.
7	What white object is lying on the table in front of the man?	A rolled-up napkin.

Example 6 — 8 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'steps by a door' information within this image?	[0.00, 0.23, 0.54, 0.83]
2	What can you infer from the mentioned visual clues? Please respond briefly.	The door is on the basement level.
3	I need assistance finding the 'a dog walking toward a door' in this image. Could you provide me with the corresponding region coordinates [xmin, ymin, xmax, ymax]?	[0.40, 0.50, 0.67, 0.86]
4	I need assistance finding the 'a shaggy dog by an entrance' in this image. Could you provide me with the corresponding region coordinates [xmin, ymin, xmax, ymax]?	[0.36, 0.52, 0.67, 0.95]
5	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'wet fur on a dog' information within this image?	[0.38, 0.48, 0.62, 0.89]
6	What text is visible on the blue sign attached to the glass door?	Waiting Room
7	Describe the clothing of the person walking up the stairs in the center background.	A brown jacket and white pants.
8	What is the person on the far left carrying?	A white bag with red text.

Example 7 — 7 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'a knife on a cutting board' information within this image?	[0.73, 0.37, 0.95, 0.89]
2	What can you infer from the mentioned visual clues? Please respond briefly.	items were cut on the board
3	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'alcohol bottles on a table' information within this image?	[0.51, 0.03, 0.76, 0.55]
4	What text labels appear at the bottom of the left and right panels respectively?	EXPECTATIONS and REALITY
5	Describe the scene and people in the left panel labeled 'EXPECTATIONS'.	A man in a suit and a woman in a patterned dress are standing close together, looking at each other and smiling.
6	What specific action is being performed with the fruit in the right panel?	A hand is squeezing a lime wedge into a plastic cup.
7	What types of citrus fruit are visible on the cutting board?	Lemon and lime wedges.

Example 8 — 7 turns on one image

#	question	answer
1	Please identify the area in this image where 'a large white plastic bag' and give me the region coordinates [xmin, ymin, xmax, ymax].	The region likely refers to the white jumpsuit worn by the person on the left, as no large white plastic bag is clearly visible.
2	What can you infer from the mentioned visual clues? Please respond briefly.	the robbers are collecting items from the hostages
3	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'two men dressed the same, with white masks, hats, and sunglasses at a bank' information within this image?	[0.00, 0.03, 0.33, 1.00]
4	What object is the woman in the blue patterned top holding up in her hand?	A silver flip phone.
5	What color is the strap visible over the shoulder of the person in the white jumpsuit?	Green.
6	What structure is visible in the background behind the crowd of people?	Vertical metal bars.
7	Describe the appearance of the man on the far right edge of the image.	He has long dark hair and a beard, and is wearing a dark suit.

Example 9 — 6 turns on one image

#	question	answer
1	Please identify the area in this image where 'floating candels near the ceiling' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.23, 0.01, 0.69, 0.10]
2	What can you infer from the mentioned visual clues? Please respond briefly.	This appears to be a place with magic capabilities.
3	Please identify the area in this image where 'two people looking at each intently' and give me the region coordinates [xmin, ymin, xmax, ymax].	[0.12, 0.10, 0.26, 0.74], [0.27, 0.16, 0.37, 0.68]
4	What text is visible on the blue book lying on the table near the girl on the right?	HOGWARTS
5	What items are placed on the table in front of the students?	Silver goblets, plates, and platters of food.
6	Describe the two characters sitting on the right side of the table in the foreground.	A boy with red hair and a girl with wavy brown hair, both wearing dark robes.

Example 10 — 5 turns on one image

#	question	answer
1	Could you locate and provide the region coordinates ([xmin, ymin, xmax, ymax]) for the 'wine glasses' information within this image?	[0.01, 0.58, 0.27, 1.00]
2	What can you infer from the mentioned visual clues? Please respond briefly.	this is a fancy dinner.
3	What is the man on the far right holding in his hand?	He is holding a wine glass.
4	What is the man on the left wearing?	He is wearing a beige zip-up hoodie over a blue t-shirt.
5	What food item is visible on the white plate in the foreground?	A slice of yellow cake, possibly cheesecake.

Downloads last month: 16

Total file size:

16.7 GB

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

sherlock_cleaned

How this was cleaned

answer vs org_answer