# EvoChart-QA Benchmark

This repository contains the benchmark for the paper:

**EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding**

Accepted at AAAI 2025.

[Homepage](https://github.com/MuyeHuang/EvoChart) | [arXiv](https://arxiv.org/abs/2409.01577)

## Evaluation

The evaluation of the generated answers should follow the criteria below.

## Evaluation Standards

We employ a combination of "Strict" and "Flex" approaches for a fair evaluation of answer correctness. Questions are labeled as either "Strict" or "Flex" to indicate which evaluation method should be used.

*   **Strict**: For questions with a definite answer, such as a numerical or textual value explicitly labeled on the chart. This approach uses exact string matching for evaluation.
*   **Flex**: For questions that require estimation of values, such as for unlabeled numerical data. This approach allows for a 5% tolerance when evaluating the correctness of the answer.