Blog & Articles
Insights from the Rosary Labs team
Part 1: Evaluating VLMs accuracy on P&IDs extraction - Metrics for Quantitative Evaluation - Rosary Labs
Improving VLMs for P&ID equipment tag extraction requires balancing results of False Positives against False Negatives. Through simple reasoning, we arrived at using F1 score among other metrics such as Accuracy, Precision and Recall with an inclination to correct for precision being the preferred metrics for quantitative evaluation.
January 22, 2026•Law May, Co-founder of Rosary Labs
Can Vision-Language Models Reliably Extract Equipment Tags from P&IDs?
Testing state-of-the-art Vision Language Models (Claude Opus 4.5, Gemini 3 Pro, GPT 5.2, and more) for P&ID asset tag extraction accuracy.
January 4, 2026•Rosary Labs Team
Why does OCR fail on P&ID Extraction?
Traditional OCR tools struggle with P&ID data extraction due to rotated text, spatial complexity, overlapping lines, and mixed formatting. Learn how modern AI approaches achieve 95% accuracy.
December 15, 2024•Rosary Labs Team