reading note for "Hierarchical Matching of Deformable Shapes"

keys: CVPR 2007, shape match, shape tree

this paper presents a method for matching deformable shapes, or more precisely, -- curves. For one curve, they first generate a tree structure called shape tree. The shape tree is generated as follow:
1) the root node stores the start point p_s, end point p_e, and middle point p_m of the curve, the stores the relative coordinates of p_m in the coordinate system defined by p_s and p_e;
2) then for each sub-curve, (e.g. p_s to p_m and p_m to p_e), define and store the relative coordinates as (1). This is done recursively until the sub-curve contains only one point.

to calculate the match cost between another curve "q" to the first curve p, it is evaluated as a recursive form:
MatchCost(q,p) = min_{b} { MatchCost(q_s -> q_b, p_s -> p_m) + MatchCost(q_b -> q_e, p_m -> p_e) + difference of the relative coordinate of b and m}, where b is a point on curve q.

This could be evaluted using dynamic programming.

This method could also be used to find a deformable shape in a cluttered image.

This disadvantage is : this method is very slow, about 50-100 times slower than shape context

reading note for "Wavelet Importance Sampling: Efficiently Evaluating Products of Complex Functions"

keys: siggraph 2003, importance sampling

This paper presents an idea for importance sampling on a function which is represented in a wavelet form. Besides, presents an idea for importance sampling on the product of two functions both represented in a wavelet form, without fully evaluting the wavelet product ( but still needs to partially evaluate the wavelet product to determine the distribution of sample points).

The idea is interesting, a figure from the paper could illustrate it

reading note for "An efficient caching-based rendering of translucent materials"

This paper uses irradiance caching for acceleration of radiance evaluation of the dipole translucent model. In previous works, one needs to evaluate the outgoing irradiance of samples on 3d surfaces, whose number is usually very large. They observe that the irradiance is rather smooth on the surface, so that it is very suitable to use irradiance caching. They have derived the gradients of irradiance values and proposed a sampling scheme.

reading note for Radiance Caching for Participating Media

a method using "irradiance caching" for participating media.
In single scattering, one need first to sample irradiance (or radiance map) at discrete samples in the media. Due to the observation that "irradiance/radiance map" is smooth in the space, we could only sample sparsely in the space, the irradiance/radiance map at an arbitrary point could be evaluted by interpolating from nearby samples. To evaluate the outgoing radiance, we need to do an integration along the view path, this step is almost unchanged comparing to previous methods.

keys: gradient, sampling scheme, siggraph 2008, TOG 2008

list not labeled emails in gmail

In gmail, you could give emails a label, and could further search the emails using the label.
when search emails using "label:apple", then it will give the emails with label "apple".
However, it seems that gmail does not provide an intuitive way to search for unlabeled emails. A way to do it is to search emails that exclude all labels, e.g.: "-label:apple -label:orange -label:banana". But it is not convinient. I suggest gmail could support search using "unlabel" or "labeled:false".