ChartQA

Overview

ChartQA is a benchmark designed to evaluate question-answering capabilities over charts and data visualizations. It tests both visual reasoning and logical understanding of various chart types including bar charts, line graphs, and pie charts.

Task Description

• Task Type: Chart Question Answering

• Input: Chart image + natural language question

• Output: Single word or numerical answer

• Domains: Data visualization, visual reasoning, numerical reasoning

Key Features

• Covers diverse chart types (bar, line, pie, scatter plots)

• Includes both human-written and augmented test questions

• Requires understanding of chart structure and data relationships

• Tests both visual extraction and logical reasoning abilities

• Questions range from simple data lookup to complex reasoning

Evaluation Notes

• Default evaluation uses the test split with two subsets:

  • human_test: Human-written questions

  • augmented_test: Automatically generated questions

• Primary metric: Relaxed Accuracy (allows minor variations in answers)

• Answers should be in format “ANSWER: [ANSWER]”

• Numerical answers may have tolerance for rounding differences

Properties

Property	Value
Benchmark Name	chartqa
Dataset ID	lmms-lab/ChartQA
Paper	N/A
Tags	Knowledge, MultiModal, QA
Metrics	relaxed_acc
Default Shots	0-shot
Evaluation Split	test

Data Statistics

Metric	Value
Total Samples	2,500
Prompt Length (Mean)	224.33 chars
Prompt Length (Min/Max)	178 / 352 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
human_test	1,250	220.17	178	352
augmented_test	1,250	228.49	186	293

Image Statistics:

Metric	Value
Total Images	2,500
Images per Sample	min: 1, max: 1, mean: 1
Resolution Range	184x326 - 800x1796
Formats	png

Sample Example

Subset: human_test

{
  "input": [
    {
      "id": "c75439e0",
      "content": [
        {
          "text": "\nHow many food item is shown in the bar graph?\n\nThe last line of your response should be of the form \"ANSWER: [ANSWER]\" (without quotes) where [ANSWER] is the a single word answer or number to the problem.\n"
        },
        {
          "image": "[BASE64_IMAGE: png, ~42.9KB]"
        }
      ]
    }
  ],
  "target": "14",
  "id": 0,
  "group_id": 0,
  "subset_key": "human_test"
}

Prompt Template

Prompt Template:

{question}

The last line of your response should be of the form "ANSWER: [ANSWER]" (without quotes) where [ANSWER] is the a single word answer or number to the problem.

Usage

Using CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets chartqa \
    --limit 10  # Remove this line for formal evaluation

Using Python

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['chartqa'],
    dataset_args={
        'chartqa': {
            # subset_list: ['human_test', 'augmented_test']  # optional, evaluate specific subsets
        }
    },
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)