alexis-roche/variational_sampler 3

A python package that implements Gaussian approximation of multivariate distributions using variational sampling (intended to supersede scikits.infer).

alexis-roche/permuttest 2

A scikit for statistical permutation tests

alexis-roche/nipy 1

Neuroimaging in python

alexis-roche/random_walker 1

A collection of image segmentation algorithms based on diffusion methods

alexis-roche/register 1

Standalone image registration package - will supersede nipy.algorithms.registration

alexis-roche/scripts 1

A collection of script files in python and other languages

alexis-roche/variana 1

New implementation of variational sampling

alexis-roche/nipy-labs 0

Work-in-progress add-ons for nipy (require lapack)

alexis-roche/nipy_paper 0

Nipy frontiers paper

issue commentintel-isl/Open3D

Poisson surface reconstruction has random behavior

I can confirm that setting n_threads=1 after re-building open3d from the current git master branch fixes the reproducibility issue. Weirdly enough, the function also runs faster. Thanks again @griegler for your input.


comment created time in a month

issue commentintel-isl/Open3D

Poisson surface reconstruction has random behavior

Thank you so much for your feedback Gernot. I will do that asap and let you know if it fixes the issue for me.


comment created time in 2 months

issue openedintel-isl/Open3D

Poisson surface reconstruction has random behavior

IMPORTANT: Please use the following template to report the bug.

Describe the bug The mesh created from a point cloud using the create_from_point_cloud_poisson method is not the same across different calls to the function.

To Reproduce Please run the following Python code assuming a point cloud file data.ply.

`import sys import open3d as o3d import numpy as np import pylab as plt

ntrials = 10 if len(sys.argv) > 1: ntrials = int(sys.argv[1])

pcd ='data.ply') meshes = [o3d.geometry.TriangleMesh.create_from_point_cloud_poisson(pcd, depth=8, width=0, scale=1.1, linear_fit=False)[0]
for i in range(1 + ntrials)] vertices = np.array([m.vertices for m in meshes]) errors = np.max(np.abs(vertices - vertices[0]), (1, 2))[1:] print('Errors:', errors) plt.plot(range(ntrials), errors, 'o:') plt.title('Poisson reconstruction randomness') plt.xlabel('Trial') plt.ylabel('Deviation wrt reference') `

Expected behavior The mesh output by Poisson reconstruction should always be the same.


Environment (please complete the following information):

  • Operating system: Ubuntu 18.04.4 LTS
  • Python version: Python 3.6.9
  • Open3D version:
  • Is this remote workstation?:no
  • How did you install Open3D?: pip
  • Compiler version (if built from source):

Additional context

created time in 2 months