SVRepair: Structured Visual Reasoning for Automated Program Repair

Researchers developed SVRepair, an AI system that fixes software bugs by analyzing visual information like screenshots and flowcharts. It converts these images into structured graphs to understand the problem, then generates code patches. In tests, it achieved up to 95% accuracy on certain benchmarks, significantly outperforming methods that only read text. This approach is especially useful for bugs involving broken layouts or missing visual elements.