Intrinsic shape context descriptors for deformable shapes
read more
Citations
Multi-view Convolutional Neural Networks for 3D Shape Recognition
Geometric Deep Learning on Graphs and Manifolds Using Mixture Model CNNs
Multi-view Convolutional Neural Networks for 3D Shape Recognition
Volumetric and Multi-view CNNs for Object Classification on 3D Data
Geodesic Convolutional Neural Networks on Riemannian Manifolds
References
Distinctive Image Features from Scale-Invariant Keypoints
SURF: speeded up robust features
A performance evaluation of local descriptors
Shape matching and object recognition using shape contexts
A performance evaluation of local descriptors
Related Papers (5)
Frequently Asked Questions (18)
Q2. What are the contributions in "Intrinsic shape context descriptors for deformable shapes" ?
In this work, the authors present intrinsic shape context ( ISC ) descriptors for 3D shapes. The authors generalize to surfaces the polar sampling of the image domain used in shape contexts: for this purpose, they chart the surface by shooting geodesic outwards from the point being analyzed ; ‘ angle ’ is treated as tantamount to geodesic shooting direction, and radius as geodesic distance. The resulting descriptor is a metadescriptor that can be applied to any photometric or geometric property field defined on the shape, in particular, the authors can leverage recent developments in intrinsic shape analysis and construct ISC based on state-of-the-art dense shape descriptors such as heat kernel signatures.
Q3. What have the authors stated for future works in "Intrinsic shape context descriptors for deformable shapes" ?
The authors intend to explore these constructions in their future work.
Q4. How many linear bins were used for the construction of the ISC descriptors?
For the construction of the ISC descriptors, radius was set to 𝑅 = 20; 5 linearly spaced radial bins and 16 angular bins were used.
Q5. What is the ambiguity in the system of coordinates?
Since the system of coordinates is local, there is a rotation ambiguity: one of the edges is chosen arbitrarily to align with the plane horizontal or vertical axis.
Q6. What is the original shape context approach?
The original shape context approach was developed for planar shapes, and is based on the observation that a set of vectors connecting a point on the shape to the rest of the shape points constitutes a rich and discriminative description of the shape.
Q7. How many correct matches were found in the shape?
Correct matches were defined as matches falling within 2% of the shape intrinsic diameter off the ground truth match (including the bilaterally symmetric one).
Q8. What is the main challenge in generalizing a surface to a point?
The main challenge here comes from the fact that the grid has to be intrinsic, since (as opposed to the case of images) surfaces have generally non-trivial curvature.
Q9. How many eigenvalues and eigenvectors were used on each shape?
The authors used the cotangent weight scheme [30] to compute the first 300 eigenvalues and eigenvectors of the Laplace-Beltrami operator on each shape.
Q10. How many descriptors can be designed using the local system of coordinates?
Using the local system of coordinates, intrinsic versions of many other semi-local descriptors popular in image analysis, such as covariance descriptors, can be designed.
Q11. What is the way to map the surface to the plane?
In the limit this is equivalent to subdividing the tangent plane: flattening applies a common scaling to all surface triangles when mapping to the plane, so angle ratios on the plane will be preserved on the surface.
Q12. What was the name of the descriptor?
In [1], another physically-inspired descriptor, the wave kernel signature (WKS) was proposed as a solution to the excessive sensitivity of the HKS to low-frequency information.
Q13. How do the authors create an isometric mapping from the patch to the plane?
The former tries to create an isometric mapping from the patch to the plane by finding a planar configuration of points whose pairwise Euclidean distance are as close as possible to the pairwise geodesic distances between the surface points.
Q14. What is the main idea behind diffusion geometry?
The studies of diffusion geometry are based on the theoretical works by Berard et al. [4] and later by Coifman and Lafon [13] who suggested to use the eigenvalues and eigenvectors of the Laplace-Beltrami operator associated with the shape to construct invariant metrics known as diffusion distances.
Q15. How many points were sampled from each shape?
Points from the transformed shape were matched to the null shape and ordered using 𝐿2 distance between the corresponding descriptors.
Q16. How did the authors make the shape context approach manageable?
In order to make this description manage-∗Supported by ANR-10-JCJC -0205, † HP2C, ‡ ISF and GIF grants.able, the authors proposed to compute its distribution over a polar or log-polar system of coordinates, binning together vectors with similar length and orientation.
Q17. What is the way to map the plane?
This partition is performed as follows: the authors map the 1-ring triangles (Figure 3, top left) into the plane by partitioning the plane into angular segments (Figure 3, top right) whose angle rations are equal to the ratios of the angles in 1-ring triangles.
Q18. What is the construction of the ISC descriptor?
Their construction involves geodesic distances from one vertex to a small neighborhood bounded by R, allowing us to terminate fast marching when reaching R.Figure 6 visualizes the distance maps computed in the descriptor space from a reference point on the human shapeto the rest of the points on that shape, as well as to the points of the dog shape.