Numerical Optimization Shift

The numerical optimization shift parameter is useful for managing very large coordinate magnitudes while maintaining good numerical stability. You can pass it at construction time and turn it off (set to None) when you need the raw values (e.g., for export).

Construct with or without a shift

import numpy as np
from pchandler.core import PointCloudData
from pchandler.geometry.optimal_shift import OptimizedShift

xyz = (np.random.rand(1000, 3) * 1e3).astype(float)

# Explicitly disable shifting
pcd = PointCloudData(xyz=xyz, numerical_optimization_shift=None)

pcd = PointCloudData(xyz=xyz, numerical_optimization_shift=OptimizedShift([500, 200, 1000]))

A warning will be thrown if the shift is not feasible.

xyz = (np.random.rand(1000, 3) * 1e6).astype(float)
pcd = PointCloudData(xyz=xyz,
                     numerical_optimization_shift=OptimizedShift([500, 200, 1000]))

Exporting with Open3D ignores the shift

When exporting, a copy is made without the shift so that downstream libraries see the unshifted coordinates.

import numpy as np
from pchandler.core import PointCloudData

xyz = (np.random.rand(500, 3) + 60_000).astype(float)
pcd = PointCloudData(xyz=xyz)

try:
    # If a shift were active, to_o3d would transparently export unshifted coords
    o3d_pcd = pcd.to_o3d(as_tensor=False)
except ModuleNotFoundError:
    pass

assert not np.allclose(xyz, pcd) # PCD has been shifted so shouldn't match
assert np.allclose(xyz, o3d_pcd.points)  # But should match to the original coordinates