Download presentation
Presentation is loading. Please wait.
Published byJune Nelson Modified over 9 years ago
1
COLOR-PLUS-DEPTH LEVEL-OF- DETAIL IN 3D TELE-IMMERSIVE VIDEO: A PSYCHOPHYSICAL APPROACH Wanwin Wu et la., MONET Group UIUC ACM MM 2011 Best Paper
2
TELE-IMMERSION
3
PSYCHOPHYSICS (PP)
4
PSYCHOPHYSICAL THRESHOLDS Method of Limits Method of Adjustment Method of Constant Stimuli
5
PROBLEMS Real-time 3D model construction Multi-party communication Inter/Intra-node synchronization Computational resource-hungry Networking resource-hungry
6
COMPUTATIONAL RESOURCE BALANCING CPU Resource Spatial Resolution: Color-plus- Depth Level- of-Detail Temporal Resolution: Frame Rate
7
CONSTRUCTION OF 3D MODELS Point Cloud Polygon Modeling
8
PARAMETERS OF COLOR-PLUS-DEPTH LEVEL-OF-DETAIL (CZLoD) Detailing parameter of CZLoD: TH var (Threshold of Variance) - “Recursively refining bisection until the variance within every polygon (triangle) is less than TH var ” - TH var decides the size of triangles - TH var decides both texture (color) and spatial resolution Degradation of CZLoD: DR (Degradation Ratio) - F i : the i th 3D frame - N x (F i ): number of vertices in F i if TH var = x
9
GOALS Finding the PP thresholds in tele-immersive videos - JNDG (Just Noticeable DeGradation) - JUADG (Just UnAcceptable DeGradation) Utilize the thresholds on constructing a real-time adaptation scheme for resource balancing - QoS Monitor - Decision Engine - Variance Calculation
10
GOAL 1: FINDING PP THRESHOLDS Physical stimuli: DR Perceptual quantity: JNDG and JUADG (participant was asked if he/she could tell any difference between the two clips, and whether he/she thought any video had unacceptable quality) Experiment method: Ascending Method of Limit Video
11
FINDING PP THRESHOLDS (CONTI.) Participants: - 16 graduate students of CS UIUC - all had normal or corrected vision - 4 Indian, 3 American, 2 Chinese, 2 German, 2 Bangladeshi, 1 Mexican, 1 South African - 6 women (37.5%) and 10 men (62.5%) - 5 experts of tele-immersion (31.25%) and 11 novices (68.75%)
12
MAPPING FROM TH var TO DR Very much content-dependent The input parameter (TH var ) is manually adjusted to form a series of gradually degrading videos
13
RESULTS The average thresholds among participants and among the four videos are JNDG: 61.5%; JUADG: 81.25% Thresholds of low resolution videos are lower (easier for users to notice degradations) Influence of content is less with higher resolution The size of gray zone is 10~20%
14
GOAL 2: ADAPTATION SCHEME
15
QoS MONITOR Input: the last 3D frame displayed Analyze: DR and construction time (frame rate) Output - DR value - abnormal frame rate (FR) event: if (FR > TH h || FR < TH l )
16
MORE DETAILS ON QoS MONITOR According to [16], (TH h, TH l ) is set as (8, 12) In calculation of DR, N 0 (F i ) is periodically computed to reduce complexity DR here is hence defined as min{1, DR(F i )} The frame rates are measured within a sliding window of 5 frames
17
DECISION ENGINE Input: DR value and abnormal FR report Output: target DR and DR error (explained later) The purpose is fairly simple, actually - If FR is too low, increase it (by ∆ d ) - If FR is too high, decrease it (by ∆ u ) The nontrivial are the amounts of changes ( ∆ d and ∆ u ) The changes can follow several protocols - constant - AIMD -...
18
WHAT ABOUT THE PP THRESHOLDS? Three decision zones The adaptation zone = [JNDG-B n, JUADG+B a ] ∆ d and ∆ u are set as 0 outside the adaptation zone
19
VARIANCE CALCULATOR Input: target DR and DR error Output: TH var The same problem again: a mapping F from DR to TH var is needed Solution: least-square regression-based learning When DR error is larger then some threshold (TH err ) then the learning process is triggered With ten training points (s 0 ~s 9 ), the median residual is 0.022%
20
PERFORMANCE EVALUATION Frame rate improvement (a) w/o extra CPU stress; (b) w/ 16% CPU stress
21
PERFORMANCE EVALUATION (CONTI.) DR – TH var mapping Crowd-sourcing (Youtube) scoring
22
SUMMARY / FUTURE WORK PP thresholds do exist in tele-immersion PP thresholds can help improving resource balancing between spatial and temporal resolutions Although mapping between parameter setting and actual CZLoD is content dependent and nontrivial, a simple regression-based learning provides sufficient prediction Expect higher computing power to support real-time N 0 computation in the future More complex protocols to deal with the adjustment ( ∆ d and ∆ u ) of FR
23
REMARKS No variance shown in the experiment results? No statistical support of significance? Sampling of stimuli needs to be non-uniform
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.