chart2text_cleaned

The chart2text__x family of the ElliotVL supervised-fine-tuning pool, after VLM cleaning.

images	71,223
QA turns	307,704
answers rewritten by the cleaning pass	3,212
QA created by the cleaning pass (new_qa)	272,951 (88.7%)
shards	7

How this was cleaned

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	Which social network has the highest percentage of respondents, and what is that value?	Facebook has the highest percentage at 93%.
2	Which two platforms are tied for the third highest position with 63%?	YouTube and Twitter are both at 63%.
3	What is the percentage value for Sephora Community?	25%.
4	What does the x-axis label indicate?	Percentage of respondents.

Example 2 — 5 turns on one image

#	question	answer
1	Which country has the second-highest number of wins?	France, with 9 wins.
2	Which countries are tied with exactly 3 wins?	Denmark, The Netherlands, Sweden, and the Soviet Union.
3	What is the label on the x-axis?	Number of Oscar winning countries
4	List the countries that have won exactly 2 Oscars.	Argentina, Austria, Czechoslovakia, Germany, Hungary, Iran, and Switzerland.
5	What is the source and year indicated at the bottom right?	© Statista 2021

Example 3 — 4 turns on one image

#	question	answer
1	What are the expenditure values for each year from 2008 to 2018?	2008: 543, 2009: 588, 2010: 580, 2011: 543, 2012: 520, 2013: 507, 2014: 569, 2015: 712, 2016: 705, 2017: 639, 2018: 764.
2	What is the label on the vertical axis?	Expenditures in million U.S. dollars.
3	Which year shows the lowest expenditure and what is the value?	2013 with 507 million U.S. dollars.
4	What is the source of the chart indicated in the bottom right corner?	Statista 2021.

Example 4 — 2 turns on one image

#	question	answer
1	What is the source indicated at the bottom right of the chart?	The source is indicated as '© Statista 2021'.
2	Which activities have a respondent share between 10% and 20%?	The activities in this range are Cooking, Studying, Play console games, House cleaning, Read newspapers, Read magazines, and Handicrafts.

Example 5 — 3 turns on one image

#	question	answer
1	What are the specific median annual salary values for each category shown in the chart?	Financial services activities: 42,125 GBP; Financial & insurance average all categories: 39,569 GBP; Activities auxiliary to financial services and insurance activities: 36,089 GBP; Insurance, reinsurance, pension funding: 35,091 GBP; Real estate activities: 26,733 GBP.
2	What is the source and copyright year listed at the bottom right of the chart?	© Statista 2021
3	What are the five categories listed on the x-axis?	Financial services activities, Financial & insurance average all categories, Activities auxiliary to financial services and insurance activities, Insurance, reinsurance, pension funding, and Real estate activities.

Example 6 — 6 turns on one image

#	question	answer
1	What is the label on the vertical y-axis?	Percentage of female adolescents
2	What is the highest percentage value shown on the chart and in which year did it occur?	81.8% in 2017
3	What is the lowest percentage value shown on the chart and in which year did it occur?	64.3% in 2010
4	What copyright information is visible in the bottom right corner?	© Statista 2021
5	What text links are visible at the bottom of the chart area?	'Additional Information' on the left and 'Show source' on the right.
6	What icons are visible in the vertical sidebar on the right?	Icons for a star, bell, settings gear, share node, quotation marks, and a printer.

Example 7 — 4 turns on one image

#	question	answer
1	What are the specific percentage values for military spending to GDP for each year from 2009 to 2019?	2009: 4.89%, 2010: 4.92%, 2011: 4.84%, 2012: 4.48%, 2013: 4.05%, 2014: 3.7%, 2015: 3.48%, 2016: 3.42%, 2017: 3.31%, 2018: 3.32%, 2019: 3.41%.
2	What is the source of the data and the copyright year shown in the bottom right corner?	Statista 2021
3	What is the label on the vertical Y-axis and what is its scale range?	The label is 'Ratio of military spending to GDP' and the scale ranges from 0% to 6%.
4	Describe the trend of the data from 2010 to 2017.	The ratio shows a consistent downward trend, dropping from a peak of 4.92% in 2010 to a low of 3.31% in 2017.

Example 8 — 4 turns on one image

#	question	answer
1	What are the values for the USA, Great Britain, and France?	64, 54.9, and 31.7 respectively.
2	What are the values for Canada, The Netherlands, Italy, Russia, Japan, and China?	27.3, 23.5, 22.4, 21.7, 19.6, and 16.1 respectively.
3	What is the label on the vertical y-axis?	Number of trips (in millions)
4	What copyright information is visible in the bottom right corner?	© Statista 2021

Example 9 — 4 turns on one image

#	question	answer
1	What percentage of the public correctly identified the country leaving the EU, and what was the answer?	60% correctly identified the United Kingdom.
2	Which conservative GOP group in the House is listed, and what percentage of people identified it?	The Freedom Caucus, with 45% awareness.
3	What are the two items with the lowest awareness percentages (37%), and what are the correct answers provided?	President of France (Macron) and Unemployment rate (4%).
4	What is the source and date of the survey presented in the chart?	Pew Research Center, survey conducted June 22-25, 2017.

Example 10 — 5 turns on one image

#	question	answer
1	Analyze the potential correlation between the percentage of U.S. adults who believe climate change is affecting their local community and the geographical distribution of climate-related disasters in the United States. How might this correlation inform our understanding of public perception and response to climate change?	The survey results indicating that 62% of U.S. adults believe climate change is affecting their local community suggest a possible correlation between public perception and the incidence of climate-related disasters. For instance, regions prone to hurricanes, wildfires, or flooding may have a higher proportion of residents who perceive climate change as a local issue. To investigate this correlation, one could overlay climate disaster data from sources like FEMA or NOAA onto a heatmap of the survey respondents' locations. This analysis could reveal patterns, such as a higher concentration of respondents in coastal areas or regions with frequent extreme weather events. This correlation could have significant implications for climate change policy and communication strategies, as it may indicate that personal experience with climate-related disasters shapes public perception and motivates individuals to take action.
2	Compare the Pew Research Center survey results to other studies on public opinion and climate change. How do the findings align with or diverge from global trends in climate change awareness and concern?	The Pew Research Center survey results, indicating that 62% of U.S. adults believe climate change is affecting their local community, can be contextualized by considering other studies on public opinion and climate change. For example, the IPCC's 2018 report on Global Warming of 1.5°C found that 70% of respondents across 22 countries considered climate change a "very serious" or "somewhat serious" problem. A global survey by the University of Oxford in 2020 reported that 71% of respondents in 14 countries believed climate change to be a "very serious" or "extremely serious" threat. While the Pew Research Center survey results are consistent with these global trends, they also suggest a potential gap in perception between the United States and other countries. This divergence could be attributed to various factors, such as differences in climate change communication strategies, policy frameworks, or cultural attitudes toward environmental issues.
3	Design a follow-up survey to investigate the relationship between individuals' perceived impact of climate change on their local community and their self-reported behaviors related to sustainability and environmental conservation. What methodological considerations would be essential to ensure the validity and reliability of the survey results?	A follow-up survey could explore the link between individuals' perceptions of climate change's local impact and their self-reported sustainability behaviors. The survey could include questions on respondents' engagement in activities like recycling, reducing energy consumption, or participating in environmental activism. To ensure validity and reliability, the survey would require careful methodological considerations, such as: (1) a representative sample of the U.S. adult population; (2)leichously designed questions to minimize social desirability bias and ensure accurate self-reporting; (3) the use of validated scales to measure sustainability behaviors and climate change concern; (4) controlling for demographic variables, such as age, education, and income, to identify potential patterns or correlations; and (5) incorporating open-ended questions to gather more nuanced, qualitative insights. By examining the relationship between perceived climate change impact and sustainability behaviors, the survey could provide valuable insights into the effectiveness of climate change communication strategies and the role of personal experience in shaping pro-environmental behaviors.
4	What are the specific response categories and their corresponding percentages shown in the pie chart?	The chart displays four categories: 'A great deal' (22%), 'Some' (39%), 'Not too much/ not at all' (38%), and 'No answer' (1%).
5	What is the source and date of the data presented in the graphic?	The data is from a survey conducted Oct. 1-13, 2019, titled 'U.S. Public Views on Climate and Energy' by the Pew Research Center.