add farthest point sampling for points sampler, and modify the uniform sampler for the meshes #253
Conversation
|
Hola @hansen7, Thank you so much for the contribution! I'm leaving a few comments in the P.R. review. Pd. I'm glad to hear that you are interested in making more commitments, don't hesitate to ask anything. |
pyntcloud/samplers/mesh.py
Outdated
| u[is_a_problem] = 1 - u[is_a_problem] | ||
| v[is_a_problem] = 1 - v[is_a_problem] | ||
| u = np.random.uniform(low=0., high=1., size=(self.n, 1)) | ||
| v = np.random.uniform(low=0., high=u, size=(self.n, 1)) |
There was a problem hiding this comment.
I really like this change!, it really simplify things. However I think that in order to maintain the expected¿? behaviour you should change:
v = np.random.uniform(low=0., high=1 - u, size=(n, 1))Right?
There was a problem hiding this comment.
oh, yes, sorry for the typos, I was meant to write 1-u
|
In addition to the above comments, please follow the steps described in: https://pyntcloud.readthedocs.io/en/latest/samplers_dev.html#let-pyntcloud-know-about-your-sampler In order to make your new sample available and easily accesible for everybody |
|
haha no worries, is there other places for development, I am looking forward to more contributions |
Hi, Castro
I have added farthest point sampling into the points sampler, which is an algorithm applied to the cutting-edge research, see here for the relevant paper, and here for the description of the algorithm.
And I think it is probably better to sample
u, vuniformly from[0, 1]and[0, u]respectively in the mesh sampler. Because to ensure the sampled points are within that face, it requires not onlyu + v <= 1but alsou >= 0andv >= 0, but for the original procedure, which is sampled from Gaussian then truncation, it did not resolve the issues with negative sampling results. Plus it is more concise now.I am very interested to make more commitments since it intersects with my research projects during PhD, I think next step is to develop the Poisson Disk Sampling as discussed here: #239, and is it okay for me to develop a new part for the normals/curvatures calculations(directly from the points) and with reconstruction on the doc as well.
Thanks.