Jack Pinches INFO410 & INFO350 S2 2015 INFORMATION SCIENCE Geometric Modelling 2.
Presented by Matthew Cook INFO410 & INFO350 S2 2015 INFORMATION SCIENCE Paper Discussion: Dynamic 3D...
-
Upload
melvin-marshall -
Category
Documents
-
view
213 -
download
0
Transcript of Presented by Matthew Cook INFO410 & INFO350 S2 2015 INFORMATION SCIENCE Paper Discussion: Dynamic 3D...
Presented by Matthew Cook
INFO410 & INFO350 S2 2015
INFORMATIONSCIENCE
Paper Discussion:Dynamic 3D Avatar Creationfrom Hand-held Video Input
A Ichim, S Bouaziz & M Pauly
INFO410 S2 2015 PAPER DISCUSSION SLIDE 2INFORMATION SCIENCE
Overview Pipeline for creating rigged, personalised face from hand-
held video Makes use of
Feature tracking Multi-view stereo depth estimation Blendshape animation/modelling
New contributions Factor out environment light to create face albedo
map Build blendshape rig from single camera video input
INFO410 S2 2015 PAPER DISCUSSION SLIDE 3INFORMATION SCIENCE
Overview - Stages
INFO410 S2 2015 PAPER DISCUSSION SLIDE 4INFORMATION SCIENCE
Static Modelling From mobile phone video input, generate 2 meshes
High res model of face detail from neutral pose Low res blendshape model for animating expressions
Later will create one blendshape pose per facial expression
INFO410 S2 2015 PAPER DISCUSSION SLIDE 5INFORMATION SCIENCE
Static Modelling Point cloud
generated with ‘Structure from Motion’
Applies features of captured face to a ‘Geometric Morphable Model’
INFO410 S2 2015 PAPER DISCUSSION SLIDE 6INFORMATION SCIENCE
Static Modelling The user is asked to mark
complex ridges for increased accuracy
Eyes are created simply by applying the detected iris colour to an existing mesh
INFO410 S2 2015 PAPER DISCUSSION SLIDE 7INFORMATION SCIENCE
Albedo Map An albedo map is a texture map of just the diffuse
reflections of a surface i.e. it tries to eliminate the effect of environment light
INFO410 S2 2015 PAPER DISCUSSION SLIDE 8INFORMATION SCIENCE
Dynamic Modelling Could stop at static
modelling and include only ‘virtual’ emotions in virtual environment
But what if you want to use the user’s own expressions?
INFO410 S2 2015 PAPER DISCUSSION SLIDE 9INFORMATION SCIENCE
Blendshapes A mesh of vertices which are weighted to vertices in
another mesh As vertices from the blendshape mesh are moved, the
attached vertices from the target are moved, according to their weights
Used to smoothly animate ‘organic’ motions
http://area.autodesk.com/tutorials/dragon_slayer
INFO410 S2 2015 PAPER DISCUSSION SLIDE 10INFORMATION SCIENCE
Blendshapes The blendshape models are generated using a
combination of feature tracking and model-fitting As with point registration, users can manually adjust the
detected points
INFO410 S2 2015 PAPER DISCUSSION SLIDE 11INFORMATION SCIENCE
Detail Synthesis Additional details can be
added after animation from known information such as surface normals
INFO410 S2 2015 PAPER DISCUSSION SLIDE 12INFORMATION SCIENCE
Results
Potential Exam Questions
INFO410 S2 2015 PAPER DISCUSSION SLIDE 14INFORMATION SCIENCE
Exam Questions Describe how you can estimate the depth of a feature
using only a single video stream What is an albedo map? What is a blendshape?
INFO410 S2 2015 PAPER DISCUSSION SLIDE 15INFORMATION SCIENCE
References The Paper: http://lgg.epfl.ch/~bouaziz/pdf/Avatars_SIGG15.pdf