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	Please generate a detailed description of the chart.	Here a bar plot is titled Average ticket price Tampa Bay Lightning (NHL) games from 2005/06 to 2014/15 (in U.S. dollars). On the y-axis, Year is defined along a categorical scale starting with 05/06 and ending with 14/15. A linear scale from 0 to 40 can be seen along the x-axis, labeled Average ticket price in U.S. dollars.
2	What is the color of the bars in the chart?	Light green
3	Which season corresponds to the shortest bar, indicating the lowest average ticket price?	08/09
4	Do any of the bars extend beyond the 40 mark on the x-axis?	Yes, the bars for 05/06, 06/07, and 14/15 extend slightly past the 40 mark.

Example 2 — 4 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	The image displays a line chart titled 'National Football League average ticket price from 2006 to 2019 (in U.S. dollars)'. The x-axis is labeled 'Year' with tick marks from 2006 to 2018. The y-axis is labeled 'Ticket price in U.S. dollars' ranging from 0 to 100. The line graph shows a consistent upward trend, starting near $60 in 2006 and rising to slightly above $100 by 2019.	National Football League average ticket price from 2006 to 2019 (in U.S. dollars) is a line chart. There is a linear scale from 2006 to 2018 along the x-axis, marked Year. Along the y-axis, Ticket price in U.S. dollars is measured along a linear scale from 0 to 100.
2	What is the general trend of the NFL average ticket price from 2006 to 2019?	The ticket price shows a consistent upward trend, starting around $60 in 2006 and rising to slightly above $100 by 2019.	(unchanged)
3	What are the specific values marked on the y-axis?	The y-axis is marked with 0, 20, 40, 60, 80, and 100.	(unchanged)
4	What is the color of the line in the chart?	The line is grey.	(unchanged)

Example 3 — 4 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	Annual number of hospital beds in the United Kingdom (UK) from 2000 to 2019 is a bar plot. A linear scale from 0 to 250,000 can be seen on the y-axis, marked Number of beds. A categorical scale from 2000 to 2019 can be seen on the x-axis, marked Year, with asterisks noted for 2010 (), 2018 (), and 2019 ().	Annual number of hospital beds in the United Kingdom (UK) from 2000 to 2019 is a bar plot. A linear scale from 0 to 250,000 can be seen on the y-axis, marked Number of beds. A categorical scale from 2000 to 2019*** can be seen on the x-axis, marked Year.
2	What is the general trend of the number of hospital beds from 2000 to 2019?	The number of hospital beds shows a consistent downward trend over the period.	(unchanged)
3	Which years on the x-axis are marked with asterisks, and how many asterisks does each have?	2010 has one asterisk (), 2018 has two (), and 2019 has three ().	(unchanged)
4	What are the colors of the bars and the background?	The bars are light beige and the background is dark grey.	(unchanged)

Example 4 — 4 turns on one image

#	question	answer
1	Please generate a detailed description of the chart.	Household penetration rates for dog-ownership in the United States from 1988 to 2016 is a area graph. The x-axis measures Year using linear scale of range 1990 to 2015 while the y-axis plots Penetration rates using linear scale with a minimum of 0.0 and a maximum of 0.5.
2	What is the approximate penetration rate at the end of the timeline in 2016?	It is close to 0.5.
3	What color is the area filled in the chart?	Light purple.
4	How does the penetration rate change between 2005 and 2010?	It increases significantly from approximately 0.4 to nearly 0.46.

Example 5 — 5 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	The image shows a horizontal bar chart titled 'B2C e-commerce as percentage of GDP in the Middle East and Africa from 2009 to 2018'. The horizontal x-axis represents 'Percentage of gross domestic product' with values from 0.0000 to 0.0012. The vertical y-axis is labeled 'Response' and lists years from 2009 to 2018.	Here a bar plot is called B2C e-commerce as percentage of GDP in the Middle East and Africa from 2009 to 2018. The x-axis plots Percentage of gross domestic product while the y-axis shows Response.
2	What specific years are listed along the vertical axis?	2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018*	(unchanged)
3	What is the range of values displayed on the horizontal x-axis?	0.0000 to 0.0012	(unchanged)
4	What color are the bars in the chart?	Pink (or magenta)	(unchanged)
5	Which years are marked with an asterisk (*)?	2014, 2015, 2016, 2017, and 2018	(unchanged)

Example 6 — 4 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	This is an area plot titled 'Per capita consumption of fresh bell peppers in the United States from 2000 to 2019 (in pounds)'. The x-axis measures Year while the y-axis plots Per capita consumption in pounds** with a scale ranging from 0 to 12.	This is a area plot named Per capita consumption of fresh bell peppers in the United States from 2000 to 2019 (in pounds). The x-axis measures Year while the y-axis plots Per capita consumption in pounds**.
2	What are the specific years labeled as tick marks on the x-axis?	2000, 2005, 2010, 2015	(unchanged)
3	What is the general trend of the per capita consumption shown in the chart?	The consumption generally increases over the period from 2000 to 2019.	(unchanged)
4	What color is the area filled in the chart?	Dark grey or black	(unchanged)

Example 7 — 3 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	This area chart is titled 'Median annual family income in the United States from 1990 to 2019 (in U.S. dollars).' The x-axis represents the Year, with labeled ticks from 1990 to 2015, while the y-axis measures Median income in U.S. dollars, ranging from 0 to 80,000.	This area diagram is named Median annual family income in the United States from 1990 to 2019 (in U.S. dollars). The x-axis shows Year with linear scale with a minimum of 1990 and a maximum of 2015 while the y-axis measures Median income in U.S. dollars with linear scale from 0 to 80,000.
2	What is the general trend of the median annual family income shown in the chart?	The income generally increases over time, starting around $35,000 in 1990, rising to a peak near $60,000 around 2007, dipping slightly around 2010, and then rising sharply to nearly $80,000 by 2019.	(unchanged)
3	What are the labeled tick marks on the y-axis?	0, 20,000, 40,000, 60,000, 80,000	(unchanged)

Example 8 — 4 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	The image is a horizontal bar chart titled 'Retail revenue compound annual growth rate (CAGR) of the world 's fastest growing retailers from FY 2013 to 2018'. The y-axis lists businesses, and the x-axis shows the FY 2013-2018 retail revenue CAGR ranging from 0.0 to 0.6. The chart displays data for ten retailers: A101 Yeni Mağazacılık A.Ş, Action Nederland BV, Albertsons Companies Inc., Dollar Tree, Inc., JD Sports Fashion Plc, JD.com, Inc., Reliance Industries Limited/Relia..., Vipshop Holdings Limited, Wayfair, Inc., and Zalando SE.	Here a is a bar plot titled Retail revenue compound annual growth rate (CAGR) of the world 's fastest growing retailers from FY 2013 to 2018. The y-axis plots business. The x-axis plots FY 2013-2018 retail revenue CAGR as a scale of range 0.0 to 0.6.
2	List the retailers shown in the chart from top to bottom.	A101 Yeni Mağazacılık A.Ş, Action Nederland BV, Albertsons Companies Inc., Dollar Tree, Inc., JD Sports Fashion Plc, JD.com, Inc., Reliance Industries Limited/Relia..., Vipshop Holdings Limited, Wayfair, Inc., Zalando SE.	(unchanged)
3	Which retailer has the highest retail revenue CAGR according to the chart?	Reliance Industries Limited/Relia...	(unchanged)
4	Which retailer has the lowest retail revenue CAGR?	Dollar Tree, Inc.	(unchanged)

Example 9 — 5 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	This line diagram is titled Algeria : Youth unemployment rate from 1999 to 2020. The y-axis plots Youth unemployment rate using a linear scale with a minimum of 0.0 and a maximum of 0.5. The x-axis represents Year, with labeled ticks at 2000, 2005, 2010, and 2015.	This line diagram is titled Algeria : Youth unemployment rate from 1999 to 2020. The y-axis plots Youth unemployment rate using a linear scale with a minimum of 0.0 and a maximum of 0.5. Year is drawn on the x-axis.
2	What are the specific years labeled on the x-axis?	2000, 2005, 2010, and 2015.	(unchanged)
3	What is the approximate youth unemployment rate at the beginning of the chart (around 1999/2000)?	Approximately 0.5.	(unchanged)
4	What is the approximate youth unemployment rate at the end of the chart (2020)?	Approximately 0.3.	(unchanged)
5	What is the color of the data line plotted on the chart?	Yellow (or gold/orange).	(unchanged)

Example 10 — 5 turns on one image

#	question	answer (cleaned)	org_answer (pre-judge)
1	Please generate a detailed description of the chart.	This is a bar chart titled 'Estimated gross domestic product (GDP) loss from removal of all illegal immigrant workers from the United States as of September 2016, by industry (in billion U.S. dollars)'. The x-axis lists economic sectors, starting with 'Agriculture, forestry, fishing, and ...' and ending with 'Wholesale and retail trade'. The y-axis shows 'GDP loss in billion U.S. dollars' from 0 to 80. The chart features teal bars representing the loss for each sector, with Manufacturing having the highest loss (approx. 74 billion) and Mining the lowest (approx. 11 billion).	Here a bar graph is labeled Estimated gross domestic product (GDP) loss from removal of all illegal immigrant workers from the United States as of September 2016 , by industry (in billion U.S. dollars). There is a categorical scale from Agriculture, forestry, fishing, and hunting to Wholesale and retail trade on the x-axis, marked economic sector. There is a linear scale from 0 to 80 along the y-axis, marked GDP loss in billion U.S. dollars.
2	Which economic sector shows the highest estimated GDP loss?	Manufacturing	(unchanged)
3	Which economic sector shows the lowest estimated GDP loss?	Mining	(unchanged)
4	What is the approximate GDP loss for the 'Wholesale and retail trade' sector?	Approximately 65 billion U.S. dollars	(unchanged)
5	What color are the bars in the chart?	Teal or green	(unchanged)

Downloads last month: 25

Total file size:

890 MB

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

oroikon_chart_captioning_cleaned

How this was cleaned

answer vs org_answer