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

Adrian Bulat, Georgios Tzimiropoulos

2016-09-29Face Alignment regression 3D Face Alignment 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%.

Results

Task	Dataset	Metric	Value	Model
Facial Recognition and Modelling	3DFAW	GTE	4.5623	3D Face alignment
Face Reconstruction	3DFAW	GTE	4.5623	3D Face alignment
3D	3DFAW	GTE	4.5623	3D Face alignment
3D Face Modelling	3DFAW	GTE	4.5623	3D Face alignment
3D Face Reconstruction	3DFAW	GTE	4.5623	3D Face alignment

Related Papers

Language Integration in Fine-Tuning Multimodal Large Language Models for Image-Based Regression2025-07-20 $S^2M^2$: Scalable Stereo Matching Model for Reliable Depth Estimation2025-07-17 $π^3$: Scalable Permutation-Equivariant Visual Geometry Learning2025-07-17 Neural Network-Guided Symbolic Regression for Interpretable Descriptor Discovery in Perovskite Catalysts2025-07-16 Imbalanced Regression Pipeline Recommendation2025-07-16 Second-Order Bounds for [0,1]-Valued Regression via Betting Loss2025-07-16 Efficient Calisthenics Skills Classification through Foreground Instance Selection and Depth Estimation2025-07-16 Vision-based Perception for Autonomous Vehicles in Obstacle Avoidance Scenarios2025-07-16