[Paper Review] Learning Continuous Face Age Progression: A Pyramid of GANs

Hongyu Yang, Di Huang+2 more|arXiv (Cornell University)|Jan 10, 2019

Face recognition and analysis35 references4 citations

TL;DR

This paper proposes a pyramid of GANs framework that jointly optimizes face age progression for high aging accuracy and identity preservation by decoupling identity-specific and age-specific features. Using a multi-scale discriminator and adversarial training, it generates photorealistic, continuous aging sequences, achieving 99.88% and 99.98% face verification accuracy at 0.001% FAR on MORPH and CACD after ~28 and 23 years of aging, respectively.

ABSTRACT

The two underlying requirements of face age progression, i.e. aging accuracy and identity permanence, are not well studied in the literature. This paper presents a novel generative adversarial network based approach to address the issues in a coupled manner. It separately models the constraints for the intrinsic subject-specific characteristics and the age-specific facial changes with respect to the elapsed time, ensuring that the generated faces present desired aging effects while simultaneously keeping personalized properties stable. To ensure photo-realistic facial details, high-level age-specific features conveyed by the synthesized face are estimated by a pyramidal adversarial discriminator at multiple scales, which simulates the aging effects with finer details. Further, an adversarial learning scheme is introduced to simultaneously train a single generator and multiple parallel discriminators, resulting in smooth continuous face aging sequences. The proposed method is applicable even in the presence of variations in pose, expression, makeup, etc., achieving remarkably vivid aging effects. Quantitative evaluations by a COTS face recognition system demonstrate that the target age distributions are accurately recovered, and 99.88% and 99.98% age progressed faces can be correctly verified at 0.001% FAR after age transformations of approximately 28 and 23 years elapsed time on the MORPH and CACD databases, respectively. Both visual and quantitative assessments show that the approach advances the state-of-the-art.

Motivation & Objective

• Address the dual challenge of aging accuracy and identity permanence in face age progression, which are often compromised in existing methods.
• Overcome limitations of prior GAN-based approaches that focus on two-age-group transitions and fail to produce smooth, continuous aging sequences.
• Enable robust age progression under variations in pose, expression, and makeup without requiring extensive preprocessing or intra-individual longitudinal data.
• Achieve photorealistic aging effects by modeling high-level facial features at multiple scales using a pyramidal discriminator architecture.
• Develop a training scheme that simultaneously optimizes a single generator with multiple parallel discriminators to ensure stable and continuous aging progression.

Proposed method

• Employ a CNN-based generator to learn target age distributions and synthesize age-progressed faces while preserving identity-specific attributes.
• Decouple identity-specific features (e.g., facial structure) from age-specific features (e.g., wrinkles, skin texture) to independently model their evolution over time.
• Introduce a pyramidal adversarial discriminator that estimates high-level facial representations at multiple scales, enabling fine-grained modeling of aging details.
• Design a compound training critic combining pixel-level L1 loss, age-related GAN loss for age transformation, and identity-dependent loss to stabilize identity preservation.
• Implement an adversarial learning scheme with a single generator and multiple parallel discriminators to enable smooth, continuous aging sequences.
• Use only two landmark points for pupil alignment during inference, minimizing preprocessing requirements and enhancing practical usability.

Experimental results

Research questions

• RQ1 Can a GAN-based framework simultaneously achieve high aging accuracy and strong identity preservation in face age progression?
• RQ2 How effective is a multi-scale pyramidal discriminator in capturing fine-grained aging details compared to single-scale or global discriminators?
• RQ3 To what extent can the proposed method generate continuous, photorealistic aging sequences under pose, expression, and makeup variations?
• RQ4 How does the proposed adversarial training scheme with multiple parallel discriminators improve the quality and continuity of aging sequences compared to single-discriminator setups?
• RQ5 Can the method generalize to long-term age progression (e.g., 20+ years) without degradation in identity consistency or realism?

Key findings

• The proposed method achieves 99.88% face verification accuracy at 0.001% FAR on the MORPH database after approximately 28.61 years of age progression.
• On the CACD database, 99.98% of age-progressed faces are correctly verified at 0.001% FAR after about 23.09 years of aging.
• Human evaluation shows that 71.74% of observers preferred the proposed method over prior works, while only 19.28% favored existing methods.
• The model generates photorealistic aging effects with minimal ghosting artifacts and high facial detail fidelity, even under pose and expression variations.
• The method outperforms state-of-the-art approaches, including commercial tools like Agingbooth and Face of the Future, in both visual quality and quantitative verification metrics.
• The approach requires minimal preprocessing—only two landmark points for pupil alignment—making it practical for real-world deployment.