A new test set called the Celeb Twins Test Set (CTTS) has been developed to evaluate face recognition systems on monozygotic twins. This dataset, containing image pairs of 80 celebrity twin sets, is unique in its metadata for distinguishing skin marks and mirror asymmetry. Current deep CNN matchers achieve over 76% accuracy on CTTS but do not utilize these distinguishing features. The paper also explores the potential of generative AI tools like Grok, ChatGPT, and Gemini to create synthetic twin images for augmenting training datasets. AI
IMPACT This dataset could lead to more robust face recognition systems capable of distinguishing identical twins.
RANK_REASON The item is a research paper introducing a new dataset and analysis. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →