An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments EG SCA ’03...
-
date post
19-Dec-2015 -
Category
Documents
-
view
220 -
download
6
Transcript of An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments EG SCA ’03...
An Evaluation of a Cost Metric for Selecting Transitions between
Motion Segments
EG SCA ’03
Speaker: Alvin
Date: 4 October 2004
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
2
Outline
IntroductionCost FunctionOptimizationEvaluation
Cross-Validation User Study
Evaluation FormConclusion & Future Works
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
3
Problem Define
Goal Design rich behaviors. Minimize the amount of manual intervention.
Input MoCap data
Methods Use an underlying probabilistic model. Re-order the original motion.
Key Proper selection of transition points.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
4
Introduction
Transition Points The discontinuities in the motion stream.
Distance Function The cost of transitioning from one frame to
another. Parameterized though user-selected weights.
Contribution Produce a set of optimized weights that select
good transitions.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
5
Cost Function
Lee
d(vi ,vj) is the weighted distance of joint velocities v weights the velocity difference with respect to d(pi ,pj) d(pi ,pj) is the weighted difference of joint orientations.
p: global translational positions of the figure q: are the orientations of joint k expressed as
quaternions.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
6
Optimization
Input 16 different segments of captured motion. Each is several seconds long.
Select 16 good transitions 26 bad transitions By a single expert and examined by two other
viewers for approval.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
7
Optimization (cont.)
Method Constrained least-squares minimization.
The weights were non-negative and symmetric.
Solved using an active set method.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
8
Optimization (cont.)
Result The velocity term
makes little effective difference in the cost.
Most of the weights were unimportant.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
9
Evaluation
Cross-Validation Robust Generalize
User Study Quality Determination
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
10
Cross-Validation
Optimized with one set of training data deleted. The resulting weights are then used to compute t
he optimal value of a transition for the deleted data set.
The average deviation was less than one frame. The median deviation was zero frames. Compute the same optimal transition in 41 of 42
cases.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
11
User Study
26 non-expert adult.Show an example using the same model.Rate how noticeable and natural the
discontinuities were, both individually and in comparison to another motion.
Show two six-second motions; the transition points are at t=3s.
No interpolation or smoothing was done.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
12
User Study (cont.)
Influential Factors Blending
Linear InterpolationEase-inEase-out
IK routineFor each transition the global position and
orientation of the character was matched.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
13
User Study (cont.)
Divide the viewers into two equal-sized groups. After viewing each sequence, ask 6 questions
using 5-points Likert scale. The result were analyzed using two-way
between-subjects analyses of variance.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
14
Conclusion & Future Works
Not contain highly dynamic motions.Limit to 16 different sequences.Use only one performer.Not contain backward motions.Velocity is not significant.Examine other cost metrics.The effects of transition mechanisms.How people perceive the transitions.
Alivn/GAME Lab./CSIE/NDHU An Evaluation of a Cost Metric for Selecting Transitions between Motion Segments
15
Evaluation Form
論文簡報部份 完整性介紹 (3) 系統性介紹 (4) 表達能力 (3) 投影片製作 (3)
論文審閱部分 瞭解論文內容 (3) 結果正確性與完整性 (4) 原創性與重要性 (4) 讀後啟發與應用:
The resulting weights are very useful for our research. Besides, Cross-Validation is a method to make our experimental results more robust;User-Study can show that our experimental results are better than others.