lumasik/captcha-25k

这是一个合成数据集，包含25,000张生成的验证码图像，专为训练和测试OCR（光学字符识别）和计算机视觉模型而设计。数据集通过自定义Python脚本生成，具有合成数据来源。它包含12种以上的噪声过滤器、失真效果和可变字体渲染，以模拟真实世界的验证码挑战。该数据集主要用于OCR基准测试和自动化识别系统的评估，但需要注意的是，作为纯合成数据，它虽然适用于字符识别基准，但未考虑行为反机器人措施（如鼠标移动或浏览器指纹识别）。数据集结构为单个train.parquet文件，包含图像和元数据，共25,000个样本，每个样本都有对应的真实文本标签。数据集大小为61.7 MB，可通过Hugging Face的datasets库流式加载或本地训练使用。

A synthetic dataset consisting of 25,000 generated captcha images, designed for training and testing OCR and computer vision models. The dataset is generated via a custom Python script and includes 12+ types of noise filters, distortions, and variable font rendering to simulate real-world captcha challenges. It is intended for OCR benchmarking and testing automated recognition systems, but as pure synthetic data, it serves as an excellent benchmark for character recognition without accounting for behavioral anti-bot measures (like mouse movement or browser fingerprinting). The dataset is structured as a single train.parquet file with images and metadata, containing 25,000 samples with corresponding ground truth text labels. It has a size of 61.7 MB and is fully compatible with the Hugging Face datasets library for streaming or local training.