Learning a Deep Reinforcement Learning Policy Over the Latent Space of a Pre-trained GAN for Semantic Age Manipulation

TL;DR

This paper introduces a reinforcement learning approach over a pre-trained GAN's latent space to achieve high-fidelity, identity-preserving semantic age manipulation in images.

Contribution

It formulates a Markov Decision Process over the GAN latent space to learn a policy for semantic attribute editing, specifically age, with high realism and identity preservation.

Findings

Successfully generates high-fidelity age-altered images

Preserves identity during age manipulation

Outperforms existing methods in realism and control

Abstract

Learning a disentangled representation of the latent space has become one of the most fundamental problems studied in computer vision. Recently, many Generative Adversarial Networks (GANs) have shown promising results in generating high fidelity images. However, studies to understand the semantic layout of the latent space of pre-trained models are still limited. Several works train conditional GANs to generate faces with required semantic attributes. Unfortunately, in these attempts, the generated output is often not as photo-realistic as the unconditional state-of-the-art models. Besides, they also require large computational resources and specific datasets to generate high fidelity images. In our work, we have formulated a Markov Decision Process (MDP) over the latent space of a pre-trained GAN model to learn a conditional policy for semantic manipulation along specific attributes under defined identity bounds. Further, we have defined a semantic age manipulation scheme using a locally linear approximation over the latent space. Results show that our learned policy samples high fidelity images with required age alterations, while preserving the identity of the person.