Automated inspection systems utilizing computer vision technology are effective in managing traffic control devices (TCDs); however, they face challenges due to the limited availability of training datasets and the difficulty in generating new datasets. To address this, our study establishes a benchmark for cost-effective model training methods that achieve the desired accuracy using data from related domains and YOLOv5, a one-stage object detector known for its high accuracy and speed. In this study, three model cases were developed using distinct training approaches: (1) training with COCO-based pre-trained weights, (2) training with pre-trained weights from the source domain, and (3) training with a synthesized dataset mixed with source and target domains. Upon comparing these model cases, this study found that directly applying source domain data to the target domain is unfeasible, and a small amount of target domain data is necessary for optimal performance. A model trained with fine-tuning-based domain adaptation using pre-trained weights from the source domain and minimal target data, proved to be the most resource-efficient approach. These results contribute valuable guidance for practitioners aiming to develop TCD models with limited data, enabling them to build optimal models while conserving resources.

Understanding unfamiliar codebases is essential yet challenging in software development. Visual aids such as call graphs and control flow graphs can help, but often lead to information overload and limited interactivity. Meanwhile, LLM-based code assistants provide accessible natural language explanations that reduce cognitive barriers, but lack spatial context for code navigation. We posit that integrating these two complementary approaches can address their respective limitations. To validate this integration, we introduce CodeVoyager, a tool that combines LLM with interactive visual aids to support more effective code comprehension. We first conducted an exploratory study (n=11) to assess the tool’s potential and identify areas for refinement. Following iterative refinement, we evaluated the enhanced tool against a widely used chat-based code assistant in a within-subjects study (n=16). Results showed that CodeVoyager improved code comprehension and increased user trust. These improvements were achieved by enabling seamless interaction between textual explanations and visual code exploration, mirroring how developers naturally discuss code. This work contributes to visual-LLM integrated developer tools through (1) a novel integration approach mirroring natural code discussion, (2) empirical evidence of improved comprehension and trust, and (3) design implications for multimodal code comprehension systems.