mintar
commented
2 years ago Damn, when I developed the symmetry handling, I knew somebody would try to model a sphere at some point, but I hoped it wouldn't be so soon. Spheres are not really well supported at the moment. :-(
The computation of symmetries happens in this function:
What it does is that it multiplies each discrete symmetry with each continuous symmetry. It does not multiply a continuous symmetry with another continuous symmetry. That is how the original code from the BOP toolkit did it, and I didn't fix it. It works for all objects except spheres. This is because there's no geometrical body with 2 continuous symmetries, and the sphere is the only body with 3 continuous symmetries.
You can however fake it: You simply pre-compute two of the continuous symmetries and put them into discrete_symmetries. I've written a script for you that does this:
#!/usr/bin/env python3
from math import sin, cos, pi
import numpy as np
import simplejson as json
DISCRETE_STEPS_COUNT = 64
def rpy_to_transformation_matrix(roll=0.0, pitch=0.0, yaw=0.0):
R = np.zeros((4, 4))
R[0, 0] = cos(yaw) * cos(pitch)
R[0, 1] = -sin(yaw) * cos(roll) + cos(yaw) * sin(pitch) * sin(roll)
R[0, 2] = sin(yaw) * sin(roll) + cos(yaw) * sin(pitch) * cos(roll)
R[0, 3] = 0.0
R[1, 0] = sin(yaw) * cos(pitch)
R[1, 1] = cos(yaw) * cos(roll) + sin(yaw) * sin(pitch) * sin(roll)
R[1, 2] = -cos(yaw) * sin(roll) + sin(yaw) * sin(pitch) * cos(roll)
R[1, 3] = 0.0
R[2, 0] = -sin(pitch)
R[2, 1] = cos(pitch) * sin(roll)
R[2, 2] = cos(pitch) * cos(roll)
R[2, 3] = 0.0
R[3, 0] = 0.0
R[3, 1] = 0.0
R[3, 2] = 0.0
R[3, 3] = 1.0
return R
symmetries_discrete = []
for roll in np.arange(DISCRETE_STEPS_COUNT) / DISCRETE_STEPS_COUNT * 2.0 * np.pi:
for pitch in np.arange(DISCRETE_STEPS_COUNT) / DISCRETE_STEPS_COUNT * 2.0 * np.pi:
R = rpy_to_transformation_matrix(roll=roll, pitch=pitch)
symmetries_discrete.append(list(R.flatten()))
model_info_dict = {
"symmetries_discrete": symmetries_discrete,
"symmetries_continuous": [{"axis": [0, 0, 1], "offset": [0, 0, 0]}],
"align_axes": [{"object": [0, 1, 0], "camera": [0, 0, 1]}, {"object": [0, 0, 1], "camera": [0, 1, 0]}]
}
with open("model_info.json", "w") as fp:
json.dump(model_info_dict, fp, indent=4)Another thing that is confusing me is a tennis ball has a logo on only one side. How does that affect the model_info.json?
I think you only have two options:
- Don't specify any symmetries at all.
- Specify the full 3 continuous symmetries.
If you strictly follow the definition of symmetries, then the logo and the rest of the texture makes your object non-symmetrical. I believe however that DOPE will do much better if you go with option 2.
rahulram70
TontonTremblay
WilkoLu
Hello.
I'm trying to properly define the symmetry for a tennis ball. Currently, I've stated nothing for the discrete symmetry. However, I defined continuous symmetry like so [{"axis": [0, 0, 1], "offset": [0, 0, 0]}, {"axis": [1, 0, 0], "offset": [0, 0, 0]}, {"axis": [0, 1, 0], "offset": [0, 0, 0]}].
I didn't know how to define align axes, so I left it the same as the one from the cylinder object "align_axes": [{"object": [0, 1, 0], "camera": [0, 0, 1]}, {"object": [0, 0, 1], "camera": [0, 1, 0]}].
Another thing that is confusing is a me is a tennis ball has a logo on only one side. How does that affect the model_info.json?
Any help would be appreciated, thank you in advance.