HTML5 <canvas> Visual Bugs Dataset
收藏资源简介:
该数据集由阿尔伯塔大学的ASGAARD实验室创建,旨在评估视觉语言模型(VLMs)在检测HTML5 <canvas>应用中的视觉错误的能力。数据集包含100张截图,其中80张为注入视觉错误的截图,20张为无错误的截图,涵盖了四种视觉错误类型(布局、渲染、外观和状态)。数据集的创建过程包括从20个开源HTML5 <canvas>应用中收集截图,并手动注入视觉错误。该数据集的应用领域主要是软件测试,特别是针对HTML5 <canvas>应用的视觉错误检测,旨在解决传统测试工具无法有效检测<canvas>应用视觉错误的问题。
This dataset was created by the ASGAARD Lab at the University of Alberta, aiming to evaluate the capability of Vision-Language Models (VLMs) in detecting visual errors within HTML5 <canvas> applications. It consists of 100 screenshots in total, including 80 screenshots injected with visual errors and 20 error-free ones, covering four types of visual errors: layout, rendering, appearance, and state. The dataset was developed by collecting screenshots from 20 open-source HTML5 <canvas> applications and manually injecting visual errors. Its main application domain is software testing, particularly visual error detection for HTML5 <canvas> applications, which is designed to solve the problem that traditional testing tools cannot effectively detect visual errors in <canvas> applications.




