Researchers have developed a new method for measuring fabric thread density using smartphone photos taken through a pick glass, aiming for a $\pm$5% accuracy tolerance. They benchmarked several AI models, including Molmo2-4B and GPT-6.1-sol, against human annotators and a spectral counter (PG-Lattice). While no method achieved the desired accuracy, the tuned Molmo2-4B model performed comparably to GPT-6.1-sol, both outperforming non-expert human counts. The study highlights that image quality, particularly reliability across multiple photos, is crucial for accurate thread counting, and current generative enhancements do not consistently improve results. AI
IMPACT Highlights limitations of current AI models in precise visual measurement tasks and the importance of image quality for reliability.
RANK_REASON Academic paper detailing a new method and benchmarking AI models on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →