def plot_errors(self):
""" Display plots of error distributions"""
import matplotlib.pyplot as plt
fig, axs = plt.subplots(2, 1, tight_layout=True)
errors, valid = tables.reprojection_error(self.reprojected,
self.point_table)
errors, valid = errors.ravel(), valid.ravel()
inliers = self.inliers.ravel()
outliers = (valid & ~inliers).ravel()
axs[0].scatter(x=np.arange(errors.size)[inliers],
y=errors[inliers],
marker=".",
label='inlier')
axs[0].scatter(x=np.arange(errors.size)[outliers],
y=errors[outliers],
color='r',
marker=".",
label='outlier')
axs[1].hist(errors[valid],
bins=50,
range=(0, np.quantile(errors[valid], 0.999)))
plt.show()
def reprojection_tables(calib, inlier_only=False):
point_table = calib.point_table
if inlier_only:
point_table = point_table._extend(valid=calib.inliers)
error, valid = tables.reprojection_error(calib.projected, point_table)
def f(axis):
return reprojection_statistics(error, valid, calib.inliers, axis=axis)
return sum_axes._map(f)
def reject_outliers(self, threshold):
""" Set outlier threshold """
errors, valid = tables.reprojection_error(self.reprojected, self.point_table)
inliers = (errors < threshold) & valid
num_outliers = valid.sum() - inliers.sum()
inlier_percent = 100.0 * inliers.sum() / valid.sum()
info(f"Rejecting {num_outliers} outliers with error > {threshold:.2f} pixels, "
f"keeping {inliers.sum()} / {valid.sum()} inliers, ({inlier_percent:.2f}%)")
return self.copy(inlier_mask = inliers)