学习姿势可控的人类重建与动态隐含领域从单一图像
IEEE transactions on visualization and computer graphics
|February 7, 2024
概括
这项研究引入了一种创新的方法,可以从单个图像中创建动态的,可控制姿势的3D人体模型. 这种方法可以实现现实的人体重建,适应各种姿势,推进计算机视觉能力.
科学领域:
- 计算机视觉 计算机视觉
- 三维重建的3D重建
- 机器学习 机器学习
背景情况:
- 从单个图像中生成可控制的3D人类模型是具有挑战性的.
- 现有的方法经常产生静态模型,缺乏姿势适应性.
- 对动画和虚拟现实等应用程序来说,控制人类模型姿势至关重要.
研究的目的:
- 从单个RGB图像开发一种可控制姿势的人类重建方法.
- 学习一种动态隐含场,能够在多种姿势中代表人类.
- 为了克服静态重建和形状姿势纠的局限性.
主要方法:
- 构建了一个特征嵌入式人类模型 (FEHM),用于跨姿势的特征传播.
- 编码了三个姿势脱的特征:全球图像,空间颜色和空间几何.
- 设计了新的隐性函数来预测动态的人类隐性场.
- 为了监督,利用了一个大规模的真实的人类化身数据集 (SimuSCAN).
主要成果:
- 在3D人体重建方面取得了最先进的性能.
- 证明有效的姿势控制和形状保存.
- 成功地将人形与姿势信息分开.
结论:
- 拟议的方法可以从单个图像中实现高保真,姿势可控的人体重建.
- FEHM和姿势脱功能对于动态的人类建模是有效的.
- 这项工作推进了单图像3D人类重建领域.
相关概念视频
01:11Deformation of Member under Multiple Loadings
When a rod is made of different materials or has various cross-sections, it must be divided into parts that meet the necessary conditions for determining the deformation. These parts are each characterized by their internal force, cross-sectional area, length, and modulus of elasticity. These parameters are then used to compute the deformation of the entire rod.
In the case of a member with a variable cross-section, the strain is not constant but depends on the position. The deformation of an...
In the case of a member with a variable cross-section, the strain is not constant but depends on the position. The deformation of an...
01:12Modeling and Similitude
Scaled modeling is a fundamental technique in engineering, enabling the study of large and complex systems by creating smaller, manageable replicas that recreate critical characteristics of the original. In hydrology and civil infrastructure, for example, scaled models of dams help analyze water flow, turbulence, and pressure. This method allows for accurate predictions of real-world behavior within a controlled environment, significantly reducing the cost and time involved in full-scale...