A state-aware, hierarchical deep learning framework for automated visual glitch detection in games
Year: 2025
Type: article
Abstract: Visual anomalies in video games can degrade user experience and impact overall software quality, highlighting the need for scalable methods within modern quality assurance (QA) pipelines. Manual testing remains resource-intensive and difficult to scale, while existing AI-based approaches often struggle to generalize across diverse rendering styles and gameplay scenarios. This paper presents a hierarchical visual anomaly detection framework that integrates game state information to enhance contextual awareness and detection accuracy. A synthetic data generation pipeline is introduced to create high-fidelity, game-specific training samples that capture the visual characteristics and edge cases of individual titles. Human-in-the-loop mechanisms support the identification of challenging scenarios and the definition of functional test conditions suitable for continuous integration workflows. The system operates continuously during production, enabling real-time detection of rendering anomalies without interfering with gameplay. The proposed framework is evaluated across three commercial game titles, demonstrating its effectiveness and adaptability. It comprises a configurable data generation pipeline, a state-conditioned detection model, and an automated anomaly identification tool, forming a modular and extensible QA solution for interactive software systems.
Author: Ciprian Paduraru
Institution: University of Bucharest
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,200
OpenAlex ID: W7117789244