IDperturb: Enhancing Variation in Synthetic Face Generation via Angular Perturbation

2026-02-21

AI Summary

This work addresses the limited intra-class diversity in face images generated by existing identity-conditioned diffusion models, which hinders the generalization capability of face recognition systems. The authors propose a geometry-driven sampling strategy that enhances intra-class diversity of synthesized images without modifying the pre-trained diffusion model. By applying angular perturbations to identity embeddings on the unit hypersphere within a constrained region that strictly preserves identity consistency, the method introduces controlled variation while maintaining semantic fidelity. This approach is the first to incorporate angular perturbation into identity-conditioned generation and demonstrates significant improvements over existing synthetic data methods, achieving superior performance on multiple mainstream face recognition benchmarks.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Natural Language Processing: Generation

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Synthetic data has emerged as a practical alternative to authentic face datasets for training face recognition (FR) systems, especially as privacy and legal concerns increasingly restrict the use of real biometric data. Recent advances in identity-conditional diffusion models have enabled the generation of photorealistic and identity-consistent face images. However, many of these models suffer from limited intra-class variation, an essential property for training robust and generalizable FR models. In this work, we propose IDPERTURB, a simple yet effective geometric-driven sampling strategy to enhance diversity in synthetic face generation. IDPERTURB perturbs identity embeddings within a constrained angular region of the unit hyper-sphere, producing a diverse set of embeddings without modifying the underlying generative model. Each perturbed embedding serves as a conditioning vector for a pre-trained diffusion model, enabling the synthesis of visually varied yet identity-coherent face images suitable for training generalizable FR systems. Empirical results demonstrate that training FR on datasets generated using IDPERTURB yields improved performance across multiple FR benchmarks, compared to existing synthetic data generation approaches.

Problem

• synthetic face generation
• intra-class variation
• face recognition
• identity-consistent
• data diversity

Innovation

• angular perturbation
• identity embedding
• synthetic face generation
• intra-class variation
• diffusion model