End to End Face Reconstruction via Differentiable PnP

TL;DR

This paper presents a two-branch neural network with a differentiable PnP layer for end-to-end 3D face reconstruction and landmark detection, achieving competitive results in the ECCV 2022 challenge.

Contribution

It introduces a novel two-branch network architecture combined with a differentiable PnP layer for improved face reconstruction and landmark detection.

Findings

Achieved top performance in the ECCV 2022 WCPA Challenge.

Outperformed existing methods on the MVP-Human dataset.

Secured 3rd place in the challenge.

Abstract

This is a challenge report of the ECCV 2022 WCPA Challenge, Face Reconstruction Track. Inside this report is a brief explanation of how we accomplish this challenge. We design a two-branch network to accomplish this task, whose roles are Face Reconstruction and Face Landmark Detection. The former outputs canonical 3D face coordinates. The latter outputs pixel coordinates, i.e. 2D mapping of 3D coordinates with head pose and perspective projection. In addition, we utilize a differentiable PnP (Perspective-n-Points) layer to finetune the outputs of the two branch. Our method achieves very competitive quantitative results on the MVP-Human dataset and wins a $3^{rd}$ prize in the challenge.