Two-stage Convolutional Part Heatmap Regression for the 1st 3D Face Alignment in the Wild (3DFAW) Challenge

TL;DR

This paper introduces a two-stage convolutional heatmap regression approach for 3D face alignment in the wild, decomposing the problem into 2D landmark detection and depth estimation, achieving top performance in the 3DFAW Challenge.

Contribution

The paper presents a novel two-stage method combining 2D heatmap regression with deep residual learning for 3D face alignment, outperforming previous approaches.

Findings

Ranked 1st in the 3DFAW Challenge

Surpassed second place by over 22%

Effective decomposition of 3D alignment into 2D and depth estimation

Abstract

This paper describes our submission to the 1st 3D Face Alignment in the Wild (3DFAW) Challenge. Our method builds upon the idea of convolutional part heatmap regression [1], extending it for 3D face alignment. Our method decomposes the problem into two parts: (a) X,Y (2D) estimation and (b) Z (depth) estimation. At the first stage, our method estimates the X,Y coordinates of the facial landmarks by producing a set of 2D heatmaps, one for each landmark, using convolutional part heatmap regression. Then, these heatmaps, alongside the input RGB image, are used as input to a very deep subnetwork trained via residual learning for regressing the Z coordinate. Our method ranked 1st in the 3DFAW Challenge, surpassing the second best result by more than 22%.