TextPecker-1.5M: A Dataset for Training and evaluating TextPecker

This repository contains the TextPecker-1.5M dataset, a new benchmark proposed in the paper "TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering".

Code and Project Page

The official implementation and project details for the TextPecker and TextPecker-1.5M dataset can be found on the GitHub repository: https://github.com/CIawevy/TextPecker

Sample Usage

You can easily load the TextPecker-1.5M dataset using the Hugging Face datasets library. The dataset is provided in two configurations: train and test

from datasets import load_dataset

# Load the full TextPecker-1.5M dataset (includes train and test splits)
dataset = load_dataset("CIawevy/TextPecker-1.5M", "default")
train_data = dataset["train"]
test_data = dataset["test"]

# Load specific split directly (more efficient for practical usage)
train_data = load_dataset("CIawevy/TextPecker-1.5M", "default", split="train")
test_data = load_dataset("CIawevy/TextPecker-1.5M", "default", split="test")

For detailed instructions on installation, model download, evaluation, and running demos, please refer to the GitHub repository.

Citation

If you find this dataset useful for your research, please cite the accompanying paper:

@article{zhu2026TextPecker,
  title   = {TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering},
  author  = {Zhu, Hanshen and Liu, Yuliang and Wu, Xuecheng and Wang, An-Lan and Feng, Hao and Yang, Dingkang and Feng, Chao and Huang, Can and Tang, Jingqun and Bai, Xiang},
  journal = {arXiv preprint arXiv:2602.20903},
  year    = {2026}
}