Computation burden during registration

[gitouni]

I run GoICP to registrate the example point cloud bunny and it takes nearly 3min... Is it a normal time? Or something went wrong? The result of running test.py is as follows:

Building Distance Transform...
REGISTERING....
INITIALIZE THE GOICP SYSTEM :: 
FINDING OUTERBNB :: 
Error*: 1817.46 (Init)
Error*: 165.804 (ICP 2.29926s)
ICP-ONLY Rotation Matrix:
   0.5741433   -0.6755550    0.4625849 
   0.8152511    0.5239115   -0.2467435 
  -0.0756646    0.5187881    0.8515467 
ICP-ONLY Translation Vector:
   0.1580152 
  -0.2472682 
   0.1960699 
Error*: 153.895
Error*: 74.5675(ICP 3.83368s)
LB=0.008700  L=3
Error*: 57.6385
Error*: 4.57052(ICP 0.509051s)
Error*: 4.57052, LB: 0.0708151, epsilon: 30.379
CLEARING THE GOICP SYSTEM :: 
TOTAL TIME :  183.44389986991882
(30379, 3) (30379, 1)
[[-0.00820469  0.00227666  0.99996367]
 [-0.00792565  0.99996578 -0.00234201]
 [-0.99993494 -0.00794464 -0.00818643]]
[ 0.217362   -0.15046362  0.0738854   1.        ]
[[-8.20468824e-003  2.27665864e-003  9.99963667e-001  2.17362004e-001]
 [-7.92565327e-003  9.99965777e-001 -2.34200733e-003 -1.50463623e-001]
 [-9.99934939e-001 -7.94464242e-003 -8.18642848e-003  7.38853965e-002]
 [ 4.63608805e-310  4.63608805e-310  4.63608805e-310  1.00000000e+000]]

[aalavandhaann]

No it's not normal. It's been a while I used go-icp for experimentation. Did you run the bunny experiment with the original c++ code,? Python takes only that much time the c++ code cosumes. Two things here. 1) check the same bunny experiment with c++, 2) what is your environment?

[gitouni]

I run go-icp in python 3.8 on Ubuntu 20.04. Actually it is the demo in this repo (test.py). The author of C++version said it takes only 20-30s but I haven't tried the C++ version.