Video based Animation Synthesis with the Essential Graph Adnane Boukhayma Edmond Boyer LJK, Universit´ e Grenoble Alpes, Inria Grenoble Rhˆ one-Alpes, France Problem Given a set of recorded elementary 4D models, generate realistic new animations that satisfy user speciﬁed constraints. Input: Video based temporally consistent Output: Novel, user-guided realistic surface animation. mesh sequences of cyclic or acyclic elementary motions. Approach Data organizing structure • Initial graph instance: Node = frame, edge = transition. • Complete directed graph: All nodes connected with a realism cost. • Essential sub-graph: Extraction of the union of shortest path trees rooted at every node. • Minimum cost graph walk: Graph search for a walk satisfying user high-level constraint. Pose parametrization Pose distance based on local deformation gradients. Pose interpolation using Poisson reconstruction of mesh deformed with geodesically interpolated deformation gradients. Interpoalted transition Generation of a transition from frame i to frame j through grad- ual frame blending with optimal segment lengths l i and l j , and temporal warps w i and w j w.r.t surface deformation cost. Contributions • The essential graph, an optimal structure for motion data organization and reuse. • Dynamic Time Warping & Variable Length Blended Segments, for more realistic interpolated motion transitions. Qualitative comparison Interpolated transition Combining dynamic time warping and variable length blended segments (a), compared to a standard interpolated transition (b). Organizing data structure Illustration of a motion graph’s (b) failure to ﬁnd a graph walk as short and costless as the essential graph (a). Quantitative comparison Organizing data structure 0 1 2 3 0 10000 20000 30000 40000 5000 (a) Motion graph Essential graph 1 2 3 4 5 6 7 8 0 10000 20000 30000 40000 5000 (b) Motion graph Essential graph 0 1 2 3 4 5 6 7 0 50000 100000 150000 200000 (c) Motion graph Essential graph 0 2 4 6 8 10 12 14 0 50000 100000 150000 200000 (d) Motion graph Essential graph Costs of all optimal paths joining all pairs of frames in the dataset for a motion graph and an essential graph. Dan dataset: (a)α =0.01, (b)α =0.1. JP dataset: (c)α =0.01, (d)α =0.1.

Essential graph Motion graph - Inria › hal-01212168 › file › Poster_3dv.pdfInitial graph instance: Node = frame, edge = transition. Complete directed graph: All nodes connected

Download PDF Report

Upload
others
View
9
Download
0

Embed Size (px)

Citation preview

Video based Animation Synthesis with the Essential GraphAdnane Boukhayma Edmond Boyer

LJK, Université Grenoble Alpes, Inria Grenoble Rhône-Alpes, France

Problem

Given a set of recorded elementary 4D models, generate realistic new animations that satisfy user specified constraints.

Input: Video based temporally consistent Output: Novel, user-guided realistic surface animation.mesh sequences of cyclic or acyclic

elementary motions.

Approach

Data organizing structure

• Initial graph instance:Node = frame, edge = transition.

• Complete directed graph:All nodes connected with a realism cost.

• Essential sub-graph:Extraction of the union of shortest pathtrees rooted at every node.

• Minimum cost graph walk:Graph search for a walk satisfying userhigh-level constraint.

Pose parametrization

Pose distance based onlocal deformationgradients.

Pose interpolation usingPoisson reconstructionof mesh deformed withgeodesically interpolateddeformation gradients.

Interpoalted transition

Generation of a transition from frame i to frame j through grad-ual frame blending with optimal segment lengths li and lj , andtemporal warps wi and wj w.r.t surface deformation cost.

Contributions

• The essential graph, an optimal structure for motion dataorganization and reuse.

• Dynamic Time Warping & Variable Length BlendedSegments, for more realistic interpolated motion transitions.

Qualitative comparison

Interpolated transition

Combining dynamictime warping and variablelength blended segments(a), compared to a standardinterpolated transition (b).

Organizing data structure

Illustration of a motiongraph’s (b) failureto find a graph walk asshort and costless as theessential graph (a).

Quantitative comparison

Organizing data structure

0

1

2

3

0 10000 20000 30000 40000 50000

(a) Motion graphEssential graph

1

2

3

4

5

6

7

8

0 10000 20000 30000 40000 50000

(b) Motion graphEssential graph

0

1

2

3

4

5

6

7

0 50000 100000 150000 200000

(c) Motion graphEssential graph

0

2

4

6

8

10

12

14

0 50000 100000 150000 200000

(d) Motion graphEssential graph

Costs of all optimal paths joining all pairs of frames in thedataset for a motion graph and an essential graph. Dan dataset:(a)α = 0.01, (b)α = 0.1. JP dataset: (c)α = 0.01, (d)α = 0.1.