ReDA: Reinforced Differentiable Attribute for 3D Face Reconstruction

2024-04-21 04:57:50

ReDA: Reinforced Differentiable Attribute for 3D Face Reconstruction

创新点：

1、针对3D人脸重建渲染成图片的约束只有2d image 的 color 属性问题，扩充为color + depth + face_parsing_mask
2、针对使用Z-buffer渲染（在cnn中不属于真正的可微分渲染）问题，提出使用CNN的操作实现soft differentiable rendering
3、针对使用3DMM作为coarse model 的问题，提出加入shape 残差，形成 free-from deformation

方法实现：

fig2展示了整体算法的流程，其中创新点free-form deformation展示在 Δ \Delta ΔS 的生成(free-form deformation), loss 约束加入了 L A R A P L_{ARAP} LARAP + R L − R m R_{L} - R_{m} RL−Rm;

fig3 展示了ReDA渲染，其中包含了较复杂的 face_parrsing_mask + color + depth；以及在不同mesh slice中的 L 1 L_{1} L1 loss约束

需检查

1、文中mean shape 是使用3DMM生成的，所以vertex number应该是 5W 点
2、文中没有展示大角度的mesh shape

ReDA: Reinforced Differentiable Attribute for 3D Face Reconstruction相关推荐

Towards High-Fidelity 3D Face Reconstruction from In-the-Wild Images
论文链接:https://arxiv.org/pdf/2003.05653.pdf Abstract 首先肯定了3DMM 从单视图图片重建人脸的巨大成功,然后点出了存在的不足--面部纹理失真,然后介绍 ...
【3d face reconstruction】综述阅读
paper:State-of-the-art in 3D face reconstruction from a single RGB image 一.methods 第一步就是要选择一个face mo ...
[论文阅读][深度学习-三维重建]Single-Shot 3D Shape Reconstruction Using Structured Light and CNN
Single-Shot 3D Shape Reconstruction Using Structured Light and Deep Convolutional Neural Networks 文章 ...
人脸对齐（十八）--Joint Face Alignment and 3D Face Reconstruction
Joint Face Alignment and 3D Face Reconstruction(2016) 本文主要提出了一种新的方法来解决任意姿态和表情的2D人脸图片的特征点定位和3D人脸重构.该方 ...
[读论文]弱监督学习的精确 3D 人脸重建：从单个图像到图像集-Accurate 3D Face Reconstruction with Weakly-Supervised Learning:From
论文地址:Accurate 3D Face Reconstruction with Weakly-Supervised Learning:From Single Image to Image Set ...
论文阅读（3）:Image-Based 3D Object Reconstruction:State-of-the-Art and Trends in the Deep Learning Era
论文标题:Image-Based 3D Object Reconstruction:State-of-the-Art and Trends in the Deep Learning Era 论文类型: ...
【论文阅读】A Point Set Generation Network for 3D Object Reconstruction from a Single Image
论文题目: A Point Set Generation Network for 3D Object Reconstruction from a Single Image (由单张图像进行三维物体重建 ...
论文阅读--Sensorless Freehand 3D Ultrasound Reconstruction via Deep Contextual Learning
Sensorless Freehand 3D Ultrasound Reconstruction via Deep Contextual Learning 基于深度上下文学习的无传感器徒手三维超声重建 ...
Large Pose 3D Face Reconstruction 文章理解
Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric 文章的一点点理解梳理写在前面因为我比较弱,在 ...

最新文章

热门文章