search

charlesq34 / pointnet

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
Other
4.71k stars 1.44k forks source link

An example or clear steps to prepare data for classification. #117

Open faisalazhar1 opened 6 years ago

faisalazhar1 commented 6 years ago

@charlesq34 Thanks for sharing the work. @karenachiketc for useful comments/solutions.

There seems to be a lot of people who are confused with regards to data preparation. I noted from the four steps of data preparation.

  1. Mesh sampling using PCL (this is uniform sampling code not the downsampling code) - convert .off to ply, read in pcl and sample it then save as *.txt (similar to PointNet++). Note only reduced point cloud will be left and polygon faces will be lost.
  2. On linux, generate *.so files from pointnet2/tf_ops/sampling/tf_sampling.py. On windows, follow the PointNet++ train file.
  3. Save 2048 points as Hdf5 or just directly train using step 2 for windows.
wusuoweima commented 5 years ago

@charlesq34 @faisal108988 I encountered a problem abuot uniformly sample

  1. I got points cloud of a CAD model from Mesh sampling using PCL . There are some points of CAD's internal structure(Internal points), but pointnet's data do not have internal points. for example, ModelNet40->airplane_0630 Can you tell me how to remove these internal points? Thank you very much!