Robust 3D Face Alignment with Multi-Path Neural Architecture Search

TL;DR

This paper introduces MONAS, a neural architecture search framework that automatically discovers effective multi-path neural networks for robust 3D face alignment across diverse poses, outperforming existing methods.

Contribution

The paper presents a novel NAS framework, MONAS, specifically designed for 3D face alignment, utilizing multi-path architectures and advanced search algorithms to improve accuracy and consistency.

Findings

MONAS achieves superior accuracy on benchmark datasets.

The multi-path architecture enhances robustness across pose variations.

Experimental results outperform existing manual-designed networks.

Abstract

3D face alignment is a very challenging and fundamental problem in computer vision. Existing deep learning-based methods manually design different networks to regress either parameters of a 3D face model or 3D positions of face vertices. However, designing such networks relies on expert knowledge, and these methods often struggle to produce consistent results across various face poses. To address this limitation, we employ Neural Architecture Search (NAS) to automatically discover the optimal architecture for 3D face alignment. We propose a novel Multi-path One-shot Neural Architecture Search (MONAS) framework that leverages multi-scale features and contextual information to enhance face alignment across various poses. The MONAS comprises two key algorithms: Multi-path Networks Unbiased Sampling Based Training and Simulated Annealing based Multi-path One-shot Search. Experimental results on three popular benchmarks demonstrate the superior performance of the MONAS for both sparse alignment and dense alignment.