ImageR: Enhancing Bug Report Clarity by Screenshots

2025-05-03

AI Summary

In bug reports, screenshots are used inefficiently—22.5% of images provide no diagnostic value, hindering issue resolution and increasing communication overhead. Method: We propose the first screenshot requirement prediction and type recommendation framework tailored to developer communication contexts. Our approach employs a multimodal classification model that jointly leverages textual semantics and image-type prior knowledge to recommend one of ten standardized screenshot categories. Contribution/Results: We release Bugzilla-Screenshot, the first publicly available dataset with fine-grained image-type annotations (6,235 bug reports). Evaluation on Mozilla projects achieves an F1-score of 0.76. A developer user study shows that 75% of participants rate the recommendations as highly valuable, reporting significant reductions in communication round-trips and measurable improvements in bug report quality and resolution efficiency.

Technology Category

• Computer Vision: Multi-modal Vision
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

In issue-tracking systems, incorporating screenshots significantly enhances the clarity of bug reports, facilitating more efficient communication and expediting issue resolution. However, determining when and what type of visual content to include remains challenging, as not all attachments effectively contribute to problem-solving; studies indicate that 22.5% of images in issue reports fail to aid in resolving the reported issues. To address this, we introduce ImageR, an AI model and tool that analyzes issue reports to assess the potential benefits of including screenshots and recommends the most pertinent types when appropriate. By proactively suggesting relevant visuals, ImageR aims to make issue reports clearer, more informative, and time-efficient. We have curated and publicly shared a dataset comprising 6,235 Bugzilla issues, each meticulously labeled with the type of image attachment, providing a valuable resource for benchmarking and advancing research in image processing within developer communication contexts. To evaluate ImageR, we conducted empirical experiments on a subset of these reports from various Mozilla projects. The tool achieved an F1-score of 0.76 in determining when images are needed, with 75% of users finding its recommendations highly valuable. By minimizing the back-and-forth communication often needed to obtain suitable screenshots, ImageR streamlines the bug reporting process. Furthermore, it guides users in selecting the most effective visual documentation from ten established categories, potentially reducing resolution times and improving the quality of bug documentation. ImageR is open-source, inviting further use and improvement by the community. The labeled dataset offers a rare resource for benchmarking and exploring image processing in the context of developer communication.

Problem

• Determining when to include screenshots in bug reports
• Identifying which visual content aids issue resolution effectively
• Reducing unnecessary images to streamline bug reporting process

Innovation

• AI model recommends relevant screenshot types
• Dataset of 6,235 labeled Bugzilla issues shared
• Open-source tool improves bug report clarity