SynthForge: Synthesizing High-Quality Face Dataset with Controllable 3D Generative Models

TL;DR

SynthForge introduces a method to generate high-quality, controllable 3D face datasets from generative models, enabling effective training for downstream tasks without relying on real data annotations.

Contribution

The paper presents a novel approach to extract 3D consistent annotations from controllable generative models, facilitating their use in downstream tasks.

Findings

Synthetic data achieves competitive performance on downstream tasks.

Method effectively extracts 3D annotations from generative models.

Demonstrates potential of synthetic data for training without real annotations.

Abstract

Recent advancements in generative models have unlocked the capabilities to render photo-realistic data in a controllable fashion. Trained on the real data, these generative models are capable of producing realistic samples with minimal to no domain gap, as compared to the traditional graphics rendering. However, using the data generated using such models for training downstream tasks remains under-explored, mainly due to the lack of 3D consistent annotations. Moreover, controllable generative models are learned from massive data and their latent space is often too vast to obtain meaningful sample distributions for downstream task with limited generation. To overcome these challenges, we extract 3D consistent annotations from an existing controllable generative model, making the data useful for downstream tasks. Our experiments show competitive performance against state-of-the-art models using only generated synthetic data, demonstrating potential for solving downstream tasks. Project page: https://synth-forge.github.io