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 — 6 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	What is the difference between the two pizzas in these images?	There is only one pizza visible in this image, which is on a red plate being held by a woman.	The pizza in the first image is on a red plate and being held by an old lady, while the pizza in the second image is on a metal counter being prepared by a woman in a blue shirt.
2	What furniture is present in the second image but not in the first image?	There is only one image provided. It shows kitchen cabinets and a counter, but no dining table or chairs are visible.	In the second image, there is a dining table and chairs, while in the first image there is no furniture visible.
3	What is the woman holding in her hands?	She is holding a red plate with a pizza on it in one hand and a white napkin in the other.	(unchanged)
4	What appliance is visible on the counter behind the woman?	A black microwave oven.	(unchanged)
5	What is sitting on top of the microwave?	A pie or baked dish in a container.	(unchanged)
6	What is the woman wearing?	She is wearing a brown long-sleeved sweater and glasses.	(unchanged)

Example 2 — 3 turns on one image

#	question	answer
1	What text is visible in the top right corner of the image?	Churchill Sea Research
2	What large white animal is being lifted in the black net?	A polar bear
3	What equipment are the two people using to lift the animal?	A white tripod frame with a red pulley mechanism

Example 3 — 7 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	What are the differences between the two images?	The image shows a single scene of a dog lying on a bed with a cellphone and pen.	One image shows a dog sitting on a bed with a cellphone and pen while the other image shows a black and white cat laying on a couch with a remote control.
2	What are the similarities between the two images?	The image shows a dog resting on a bed.	Both images show animals resting on furniture, with the first image showing a dog on a bed and the second image showing a cat on a couch.
3	What is the difference between the two images in terms of animals?	The image shows a dog lying on a bed.	In the first image, a dog is sitting on the bed while in the second image, a guinea pig is seen in a cage.
4	How do the two images differ in terms of the use of the cell phone?	A blue cell phone is lying on the bed next to the dog.	In the first image, the cell phone is lying on the bed next to the dog, while in the second image, a person is holding the cell phone in front of the guinea pig.
5	What animal is lying on the bed?	A small, fluffy dog with brown and black fur.	(unchanged)
6	What objects are lying on the bed next to the dog?	A blue mobile phone and a pen.	(unchanged)
7	What is the color of the bedding?	White.	(unchanged)

Example 4 — 6 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	How is the person in image A different from the person in image B?	The person in the image is holding ski poles.	The person in image A is holding ski poles, while the person in image B is carrying a backpack.
2	What is the difference between the skis in image A and image B?	The person is standing on the skis, not holding them.	The skis in image A are being held by the person, while the skis in image B are on the ground and not being held by the person.
3	What is the person holding in their hands?	Ski poles.	(unchanged)
4	What is the person wearing on their head?	A yellow or brown hat.	(unchanged)
5	What is visible in the background behind the person?	A snowy landscape with green bushes or vegetation.	(unchanged)
6	What color are the person's gloves?	Blue.	(unchanged)

Example 5 — 6 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	How are the pizzas different in the two images?	The image shows a single square-shaped pizza with pepperoni toppings.	In the first image, there is a square-shaped pizza with meat toppings while in the second image, there are two oddly-shaped pizzas that have been sliced into portions.
2	What is the difference in the location of the pizzas in the two images?	The pizza is sitting on a dark metal baking sheet on a stovetop.	In the first image, one pizza is on a metal tray on a stove while the other pizza is on a cooking pan on a stove. In the second image, one pizza is sitting on top of an oven and the other pizza is sitting on the stove burners.
3	What is the shape of the pizza in the image?	The pizza is square-shaped.	(unchanged)
4	What meat topping is visible on the pizza?	Pepperoni slices are visible on the pizza.	(unchanged)
5	What is the pizza resting on?	The pizza is resting on a dark grey metal baking sheet or tray.	(unchanged)
6	What kitchen appliance parts are visible in the background?	Black stovetop grates are visible on the right side.	(unchanged)

Example 6 — 5 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	What is the difference between the two bakery images?	The image shows a woman standing behind a counter with a large chocolate cake and a sign that says 'BOB'S LAST MEAL'.	In the first image, a woman is standing behind the counter with cakes for sale, while in the second image, there is only a plate of chocolate cake with a fork on the table.
2	Are there any objects that are present in both images?	There are jars or bottles visible on the shelves in the background.	Yes, there is a bottle present in both images.
3	What text is written on the white sign stuck in the cake?	The sign reads 'BOB'S LAST MEAL' and 'CHOC CAKE'.	(unchanged)
4	Describe the woman standing behind the counter.	She has blonde hair and is wearing a green shirt.	(unchanged)
5	What items are stacked on the counter to the left of the woman?	There is a stack of long bread loaves or baguettes.	(unchanged)

Example 7 — 5 turns on one image

#	question	answer
1	What route number is displayed on the destination sign of the bus?	28A
2	What are the primary colors of the bus in the foreground?	Yellow and red
3	What is visible directly behind the yellow bus?	A red bus
4	What distinctive architectural features are visible on the building in the background?	Multiple chimneys and a dome
5	What type of lighting fixtures are visible along the street?	Black, ornate street lamps

Example 8 — 4 turns on one image

#	question	answer
1	What is the woman holding to her ear?	She is holding a black smartphone.
2	What object is standing on the surface to the left of the woman?	A brown glass beer bottle.
3	Describe the chair the woman is sitting on.	It is a wooden chair with a white cushion featuring a dark floral or leaf pattern.
4	What is visible in the background?	A blurred green area, likely grass, and a stone wall on the right side.

Example 9 — 3 turns on one image

#	question	answer
1	What famous triangular building is visible in the background?	The Flatiron Building.
2	What type of numerals are used on the clock face?	Roman numerals.
3	What is the clock mounted on?	A fluted column.

Example 10 — 3 turns on one image

#	question	answer
1	What is the bird doing in the foreground?	The bird is standing on the sandy beach near the water.
2	What is visible in the background behind the bird?	The ocean with waves and a sky with orange and yellow colors, suggesting sunset.
3	What is the lighting condition in the image?	The lighting suggests it is sunset or sunrise, with a warm, golden hue.

Downloads last month: 16

Total file size:

4.79 GB

You need to agree to share your Hugging Face information to access this dataset

mimic_cgd_cleaned

How this was cleaned

answer vs org_answer