dfps-1234/echarts-synchart

ECharts-SynChart: A Synthetic Dataset for Chart Classification


This repository contains the code and dataset for the paper "ECharts-SynChart: A Large-Scale Synthetic Dataset for Chart Type Classification".

Dataset

• —Size: 10,272 images, 24 chart types.
• —Source: Generated from ECharts official examples with data augmentation (numerical perturbation, color variation, label replacement).
• —Download: https://doi.org/10.5281/zenodo.19401089 The dataset is accompanied by train/val/test CSV split files (included in this repository). The dataset is provided as a single compressed archive (echarts-synchart-images.tar.gz, ~486 MB). After downloading, extract the images; the archive contains an images/folder. Place this images/folder in the repository root (alongside README.md).

Code Structure

• —scripts/ – Python scripts for data augmentation, training, and evaluation.
• —train_classifier.py – Train ResNet50 baseline model.
• —plot_confusion_matrix.py – Generate confusion matrix on test set.
• —csv_to_jsonl.py – Convert CSV labels to JSONL format (for Qwen-VL).
• —unified_augment.py – Data augmentation for ECharts JS files.
• —generate_labels.py – Extract chart type labels from JS files.
• —merge_datasets.py – Merge multiple PNG directories into one dataset.
• —check_png.py – Detect and optionally delete blank images.
• —config.py & utils.py – Configuration and helper functions.
• —node_scripts/ – Node.js scripts for rendering ECharts JS to PNG.
• —render_echarts.js – Render a JS file to PNG using Puppeteer.
• —requirements.txt – Python dependencies.

Quick Start

1. 1.Clone this repository

  git clone https://github.com/dfps-1234/echarts-synchart.git
  cd echarts-synchart

1. 1.Install dependencies

  pip install -r requirements.txt

1. 1.Download the dataset Download the archive from Zenodo and extract it:

   tar -xzf echarts-synchart-images.tar.gz      # 解压后得到 images/ 文件夹

Place the images/folder in the root directory of this repository (alongside README.md). The default DATAROOT in trainclassifier.py is set to'.', which expects the images/folder in the repository root. The CSV files (train.csv, val.csv, test.csv) are already included in the repository – you do not need to regenerate them.

1. 1.Train the baseline model

     # Ensure DATA_ROOT in train_classifier.py points to the directory containing images/ (e.g., DATA_ROOT = '.')
   python scripts/train_classifier.py --batch_size 64 --epochs 20

1. 1.Generate confusion matrix

     # The confusion matrix script uses the same DATA_ROOT setting as train_classifier.py.
   python scripts/plot_confusion_matrix.py

Requirements

Python 3.8 or higher is required. See requirements.txt for a full list. Main packages:

• —torch >= 2.0.0
• —torchvision >= 0.15.0
• —scikit-learn >= 1.2.0
• —matplotlib >= 3.5.0
• —pandas >= 1.5.0
• —tqdm >= 4.65.0
• —Pillow >= 9.0.0
• —numpy >= 1.23.0

Citation

If you use this code or dataset in your research, please cite:

@article{li2024echarts,
  title={ECharts-SynChart: A Large-Scale Synthetic Dataset for Chart Type Classification},
  author={Li, Yunzhe},
  journal={...},
  year={2024},
  doi={10.5281/zenodo.19401089}
}

License

This project is licensed under the MIT License – see the LICENSE file for details.