|
|
AGORA: Avatars in Geography Optimized for Regression Analysis
abstract
image.png
A synthetic dataset with high realism and highly accurate ground truth, use 4240 commercially-available, high-quality, textured human scans in diverse poses and natural clothing, including 257 scans of children.
We create reference 3D poses and body shapes by fitting the SMPL-X body model (with face and hands) to the 3D scans, taking into account clothing, create around 14K training and 3K test images by rendering between 5 and 15 people per image using either image-based lighting or rendered 3D environments, taking care to make the images physically plausible and photoreal.
Extend smpl-x model better to capture children shape
other dataset
image.png
Method: Obtaining reference data
purchased high-quality textured 3D scans from:
3DPeople
AXYZ
Human Alloy
Renderpeople
选取了4240个扫描模型，包括350个独立的个体,高精度的3d扫描模型包括顶点和表面，需要将一个smpl-x模型拟合到3D扫描模型上，Fitting a SMPL-X mesh M to a scan S amounts to solving for the optimal parameters (θ, β, ψ) ；如何考虑衣服的影响：同一个人的shape参数一样
progress
image.png
使用虚拟的相机参数，渲染得到多视角的图像，使用 SMPLify-X fitting 算法得到初始化的smpl-x模型参数，为了进一步将两个3d模型进行对齐，采用：use the idea of fitting body shape under clothing:
The naked truth: Estimating body shape under clothingDetailed, accurate, human shape estimation from clothed 3D scan sequences
代价函数：
image.png
children fitting
购买的扫描模型中包含了257个小孩的3d模型，采用上述过程产生的结果不好，由于smpl-x的表达方式对于小孩并不是很鲁棒：
image.png
为了更好的拟合小孩，所以加入了 全局参数，然后一起进行优化，但是结果还是不是很自然，we take the mean infant body template from SMIL [20] and convert it to SMPL-X topology, TC ，将婴儿模型SMIL转换为SMPL-x的拓扑结构，进而让两者进行一个结合来优化。
AGORA Dataset
4240 scans spanning more than 350 unique subjects， all paired with SMPL-X fits
| Number_scans | body | hand | Face | | 3161 | | | | | 1079 | | X | X | | 4240 |
dataset split:
| Name | train | validation | Test | | Num_scans | 2930 | 259 | 1052 | | Num_images | 14529 | 1225 | 3387 |
Image-render
Camera: focal lengths 18mm, 28mm and 50mm
render-engine:unreal-engine
Env: a single Windows 10 PC with NVIDIA RTX 2080 graphics hardware
photorealistic:
image-based lighting using freely available HDR backgrounds
free and commercial 3D environments obtained from the Unreal Marketplace
对于一些坐的扫描模型添加椅子来增加真实感
other tricks.
|
本帖子中包含更多资源
您需要 登录 才可以下载或查看，没有账号？立即注册
×
|
窥视卡
雷达卡
发表于 2021-12-24 14:07
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
千斤顶
显身卡