This research introduces an innovative approach to automatically generate UML code from diagram images using multimodal large language models (MLLMs), bridging the gap between visual design and implementation.

• Creates synthetic UML activity and sequence diagram datasets for training and testing
• Leverages MLLMs to interpret visual diagrams and convert them to executable code
• Compares performance against standard methods for diagram-to-code conversion
• Demonstrates potential for automating a previously manual engineering task

For engineering teams, this advancement could significantly streamline software development workflows by automating the translation between visual designs and implementable code, reducing manual coding errors and accelerating development cycles.

Unified Modeling Language Code Generation from Diagram Images Using Multimodal Large Language Models