SMPL: A Skinned Multi-Person Linear Model (SIGGRAPH Asia 2015)

Рет қаралды 66,744

Күн бұрын

Loper, M., Mahmood, N., Romero, J.. Pons-Moll, G., Black, M.J.,
ACM Trans. Graphics (Proc. SIGGRAPH Asia), 34(6):248:1-248:16, October 2015
model: smpl.is.tue.mpg...
pdf: files.is.tue.mp...
Abstract:
We present a learned model of human body shape and pose-dependent shape variation that is more accurate than previous models and is compatible with existing graphics pipelines. Our Skinned Multi-Person Linear model (SMPL) is a skinned vertex-based model that accurately represents a wide variety of body shapes in natural human poses. The parameters of the model are learned from data including the rest pose template, blend weights, pose-dependent blend shapes, identity-dependent blend shapes, and a regressor from vertices to joint locations. Unlike previous models, the pose-dependent blend shapes are a linear function of the elements of the pose rotation matrices. This simple formulation enables training the entire model from a relatively large number of aligned 3D meshes of different people in different poses. We quantitatively evaluate variants of SMPL using linear or dual-quaternion blend skinning and show that both are more accurate than a Blend-SCAPE model trained on the same data. We also extend SMPL to realistically model dynamic soft-tissue deformations. Because it is based on blend skinning, SMPL is compatible with existing rendering engines and we make it available for research purposes.

Пікірлер: 16

@MaxwellClerkJames-f5x 6 ай бұрын

U guys will be written in the textbook. Now 2024, I am still learning u skills

@Ibormeithify 9 жыл бұрын

Fantastic work and an excellent presentation. I suppose this is based on the previous Dyna model? As a game developer, I would love to get my hands on this!

@MichaelBlackMPI 9 жыл бұрын

+Ibormeithify A first version of the model is now online. smpl.is.tue.mpg.de/ We currently support the ability to edit body shape and pose in Maya but support for Blender, etc. is coming. There are also some sample animations that can be loaded into other packages (except, currently, Unity) and played. You can't edit these though (or it will look bad if you do) as this requires a script to compute the model parameters.

@ScanlabPhotogrammetryVancouver 9 жыл бұрын

Amazing you guys!

@roxonogueira 9 жыл бұрын

I don't think I understood most of the video. I was wondering about how it would work with current 3D applications and then you mentioned maya blender and unity. Would it work with a human IK system? I didn't understand how it learns blend shapes, is it something you would have to do in house? or something uploaded to the internet to be compared to other models and animations?

@MichaelBlackMPI 9 жыл бұрын

+Pedro Nogueira, there is a version on-line that you can try smpl.is.tue.mpg.de/ The model includes all the blend shapes and the script to apply them. We are currently have only published the scripts for Maya but more is coming, including better Maya support.

@roxonogueira 9 жыл бұрын

+Michael Black thank you! I re-read my original comment and now it sounds so angry and frustrated. I was probably just excited to watch this technology and how you guys handle it! It looks incredible! Great job!!

@plavix221 9 жыл бұрын

+Michael Black I wonder what those blend shapes would look like if researchers like would use the power of deep learning. This work looks really great.

@MichaelBlackMPI 9 жыл бұрын

+plavix221 Thanks for the comment. Yes, we have thought of this but here we wanted to keep everything as simple as possible. The goal was to make the model as compatible as possible with game engines, while keeping the accuracy/realism.