Revisiting Face Recognition for Monozygotic Twins: The Celeb Twins Test Set
The Celeb Twins Test Set (CTTS) contains web-scraped image pairs for 80 sets of celebrity twins. It is the only twins test set with meta-data for twins with distinguishing skin marks and possible mirror asymmetry. Current deep CNN matchers can achieve over 76% accuracy in classifying CTTS same-person / different-person image pairs. The paper discusses the feasibility of using generative AI tools such as Grok, ChatGPT and Gemini to create images of imagined monozygotic twins.
A research paper introduces the Celeb Twins Test Set (CTTS), a face verification benchmark for monozygotic twins containing 80 celebrity twin pairs with metadata on skin marks and mirror asymmetry. Current deep CNN matchers achieve over 76% accuracy on CTTS but do not utilize skin marks or asymmetry. The paper also explores using generative AI tools like Grok, ChatGPT, and Gemini to synthesize twin images for training data augmentation.
The paper highlights that current deep CNN face matchers do not leverage skin marks or mirror asymmetry, which are key discriminative features for distinguishing identical twins. This suggests a gap in feature extraction and attention mechanisms. The proposed use of generative AI to create synthetic twin images could address data scarcity but may introduce distribution shift if not carefully validated.
Face recognition systems remain vulnerable to twin confusion, which has implications for security and identity verification. The introduction of a dedicated twins benchmark may drive development of more robust biometric systems. The mention of generative AI tools indicates growing interest in synthetic data for biometric training, potentially reducing reliance on scarce real twin datasets.
Improved twin discrimination could enhance biometric security products, reducing false accepts in high-security applications. The benchmark may become a standard evaluation for face recognition vendors. Synthetic data generation for twins could lower data acquisition costs and address privacy concerns.
Future work may focus on incorporating skin mark and asymmetry features into face recognition models, and on validating synthetic twin data for training. If generative AI can produce realistic twin images, it could enable larger-scale training and improve twin discrimination accuracy. Watch for follow-up papers reporting improved accuracy on CTTS using such approaches.