SLIDE 1
Fast Object Segmentation in Unconstrained Video Anestis Papazoglou - - PowerPoint PPT Presentation
Fast Object Segmentation in Unconstrained Video Anestis Papazoglou - - PowerPoint PPT Presentation
Fast Object Segmentation in Unconstrained Video Anestis Papazoglou and Vittorio Ferrari Outline Introduction Related Work Method Results References Introduction Video object segmentation is the task of separating foreground
SLIDE 2
SLIDE 3
Introduction
Ø Video object segmentation is the task of
separating foreground objects from the background in a video
Ø Important for a wide range of applications,
including providing spatial support for learning
- bject class models, video summarization, and
action recognition
SLIDE 4
Introduction
Ø There are two main model for segmentation:
- Require user annotation: for example, user should
annotate the object position
- Fully automatic: the only input is the input video
SLIDE 5
Introduction
Ø This paper proposes a technique for fully
automatic video object segmentation in unconstrained settings
Ø It makes minimal assumptions about the
video:the only requirement is for the object to move differently from its surrounding background in a good fraction of the video
SLIDE 6
Related Work
Ø
Object Segmentation by Long Term Analysis of Point Trajectories (T. Brox, J. Malik), ECCV 2010.
- they describe a motion clustering method
SLIDE 7
Related Work
Ø
Object Segmentation by Long Term Analysis of Point Trajectories (T. Brox, J. Malik), ECCV 2010.
– temporally consistent clusters over many frames can be
- btained best by a nalyzing long term point trajectories rather
than two-frame motion fields.
SLIDE 8
Related Work
Ø
Key-Segments for Video Object Segmentation (Y.J. Lee, J. Kim, K. Grauman), ICCV 2011.
SLIDE 9
Method
Ø The method aims to segment objects that
move differently than their surroundings.
SLIDE 10
Method
Ø The method consists of two steps:
I. Initial foreground estimation
- III. Foreground-background labelling refinement
SLIDE 11
Method
I. Initial foreground estimation
- The goal of the first stage is to rapidly produce an
initial estimate of which pixels might be inside the
- bject based purely on motion.
- The motion boundaries detected by optical flow
SLIDE 12
Initial foreground estimation
i. Optical flow estimation
SLIDE 13
Initial foreground estimation
ii. Motion Boundaries
b p
m=1−exp(−λ∥∇ ⃗
f p∥)
SLIDE 14
Initial foreground estimation
ii. Motion Boundaries
b p
θ=1−exp(−λ θmaxq∈N (δθ p , q 2 ))
SLIDE 15
Initial foreground estimation
ii. Motion Boundaries
b p={ b p
m
if b p
m>T
b p
m.b p θ
if b p
m≤T
SLIDE 16
Initial foreground estimation
- iii. Inside-outside maps
SLIDE 17
Method
- II. Foreground-background labelling refinement
➢ They formulate video segmentation as a pixel
labelling problem with two labels (foreground and background)
SLIDE 18
Method
- II. Foreground-background labelling refinement
➢ Appearance Model ( )
- The appearance model consists of two GMM over RGB
colour values,one for the foreground and one for the background.
- They are estimated automatically based on the inside-
- utside maps
- Weight of each superpixel in frame t'
- foreground:
background:
A
t
M
t
exp(−λ
A.(t−t ') 2).ri t '
exp(−λ
A.(t−t ') 2).(1−ri t ')
si
t '
SLIDE 19
Method
- II. Foreground-background labelling refinement
➢ Location Model ( )
- inside-outside maps can provide a valuable location prior to
anchor the segmentation to image areas likely to contain the
- bject, as they move differently from the surrounding region
L
t
SLIDE 20
Method
- II. Foreground-background labelling refinement
- Location Model ( )
L
t
SLIDE 21
Method
- II. Foreground-background labelling refinement
➢ Smoothness Terms
- Spatial smoothness potential
- Temporal smoothness potential
SLIDE 22
Method
- II. Foreground-background labelling refinement
➢ Smoothness Terms
SLIDE 23
Results
1) SegTrack
SLIDE 24
Results
1) SegTrack
SLIDE 25
Results
2) Youtube Objects
SLIDE 26
Results
2) Youtube Objects
SLIDE 27
SLIDE 28