Learning layered motion segmentations of video
read more
Citations
A survey of advances in vision-based human motion capture and analysis
Progressive search space reduction for human pose estimation
Chaotic invariants of Lagrangian particle trajectories for anomaly detection in crowded scenes
Bilayer Segmentation of Live Video
Track to the future: Spatio-temporal video segmentation with long-range motion cues
References
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data
Fast approximate energy minimization via graph cuts
Interactive graph cuts for optimal boundary & region segmentation of objects in N-D images
Related Papers (5)
Interactive graph cuts for optimal boundary & region segmentation of objects in N-D images
Frequently Asked Questions (9)
Q2. How long does it take to estimate the likelihood of the transformations?
The initial estimation takes approximately 5 minutes for every pair of frames: 3 minutes for computing the likelihood of the transformations and 2 minutes for MMSE estimation using LBP.
Q3. How does the algorithm account for the differences in the translations?
When determining rigidity of two transformations or clustering patches to obtain components, the authors allow for the translations to vary by one pixel in x and y directions to account for errors introduced by discretization of putative transformations.
Q4. What is the shape of a segment pi?
The shape of a segment pi is modelled as a binary matte ΘMi of size equal to the frame of the video such that ΘMi(x) = 1 for a point x belonging to segment pi (denoted by x ∈ pi) and ΘMi(x) = 0 otherwise.
Q5. What is the weight given to the contrast and the prior term?
Recall that λ1 and λ2 are the weights given to the contrast and the prior term which encourage boundaries of segments to lie on image edges.
Q6. What is the posterior probability of the model?
Given data D, i.e. the nF frames of a video, the posterior probability of the model is given byPr(Θ|D) = 1Z exp(−Ψ(Θ|D)), (3)where Z is the partition function.
Q7. What is the method to refine the estimate of the shape parameters M?
In this section, the authors describe a method to refine the estimate of the shape parameters ΘM and determine the layer numbers li using the αβ-swap and α-expansion algorithms [5].
Q8. What is the value of the pairwise potential?
As can be seen from the table, the value of the pairwise potential is small when boundaries of the segment lie on image edges (i.e. when i 6= k and gik(x,y) = 3σ).
Q9. What is the cost of assigning labels hx and hy to neighbouring points?
The cost Vx,y(hx, hy) of assigning two different labels hx and hy to neighbouring points x and y is directly proportional to Bi(x,y;Θ,D) for that frame.