search

mks0601 / V2V-PoseNet_RELEASE

Official Torch7 implementation of "V2V-PoseNet: Voxel-to-Voxel Prediction Network for Accurate 3D Hand and Human Pose Estimation from a Single Depth Map", CVPR 2018
https://arxiv.org/abs/1711.07399
MIT License
377 stars 69 forks source link

How background removal works for train data? #36

Closed jb892 closed 5 years ago

jb892 commented 5 years ago

Hi Moon,

I'm having problem about the voxelization part. I understand how you generate the refPoint for background removal. However, my implementation result doesn't seem to be correct. Here is the Point cloud: pcd_itop Resulting Voxels. voxel Why the background is not removed? What is the mechanism that controls how to crop the point cloud based on the refPoint?

mks0601 commented 5 years ago

Can you show me where is the reference point in the below figure?

jb892 commented 5 years ago

pcd2 pcd1 I cannot show it in Voxellation. I just plot the original point cloud and refpoint. The refpoint seems correct to me

mks0601 commented 5 years ago

generate_cubic_input of util.lua creates voxel grids from point cloud.

In discretize of util.lua, point cloud coordinates are discretized by pre-defined croppedSz. After that, scattering of util.lua scatter discretized coordinates in the voxel grids. Note that points whose coordinates is outside of voxel grids are removed by lower_mask and upper_mask. Did you use my code?

jb892 commented 5 years ago

No, I never used lua before. But I have read your code and try to understand what's going on. For Scattering part, it is really hard to understand. For example, the line you get the lower mask: lower_mask = (coord[1]:ge(1)):cmul(coord[2]:ge(1)):cmul(coord[3]:ge(1)). What is ge? What's going on here? Could you please explain in details what have you done in scattering part? Thanks a lot!

mks0601 commented 5 years ago

You can find the meaning of ge or le in google easily. There is an official lua torch doc and many sites https://www.google.com/search?q=lua+torch+ge&rlz=1C1NHXL_enKR711KR711&oq=lua+torch+ge&aqs=chrome..69i57.3127j0j1&sourceid=chrome&ie=UTF-8

ge means greater or equal than, and le means less or equal than.

jb892 commented 5 years ago

Cool thanks