LLMShot数据集
收藏资源简介:
LLMShot数据集由比克内特大学创建,旨在解决快照测试维护中的挑战。该数据集包含17个测试用例,每个用例都有真实的快照差异,这些差异代表真实的开发工作流程。数据集使用了一个功能丰富的iOS应用程序,具有可配置的功能标志,以创建产生真实快照差异的场景。数据集的每个记录包含参考图像、失败图像、像素级差异图像和元数据。LLMShot框架利用视觉大型语言模型来分析和分类UI变化,区分真正的回归和故意的设计修改。该数据集可用于评估LLMShot框架的性能,并为开发人员提供对UI变化的语义理解,从而减少手动审查工作量。
The LLMShot Dataset was developed by Bicknell University to address the challenges of snapshot test maintenance. It includes 17 test cases, each containing real snapshot discrepancies that represent real-world development workflows. The dataset uses a feature-rich iOS application with configurable feature flags to create scenarios that generate authentic snapshot differences. Each record in the dataset comprises reference images, failed images, pixel-level difference images, and metadata. The LLMShot framework utilizes visual large language models to analyze and categorize UI changes, distinguishing between true regressions and intentional design modifications. This dataset can be used to evaluate the performance of the LLMShot framework, and provides developers with semantic understanding of UI changes, thereby reducing manual review workload.

- 1LLMShot: Reducing snapshot testing maintenance via LLMs比克内特大学 · 2025年



