def add(R_est, t_est, R_gt, t_gt, model):
"""
Average Distance of Model Points for objects with no indistinguishable views
- by Hinterstoisser et al. (ACCV 2012).
:param R_est, t_est: Estimated pose (3x3 rot. matrix and 3x1 trans. vector).
:param R_gt, t_gt: GT pose (3x3 rot. matrix and 3x1 trans. vector).
:param model: Object model given by a dictionary where item 'pts'
is nx3 ndarray with 3D model points.
:return: Error of pose_est w.r.t. pose_gt.
"""
pts_est = misc.transform_pts_Rt(model['pts'], R_est, t_est)
pts_gt = misc.transform_pts_Rt(model['pts'], R_gt, t_gt)
e = np.linalg.norm(pts_est - pts_gt, axis=1).mean()
return e
def add(R_est, t_est, R_gt, t_gt, pts):
"""Average Distance of Model Points for objects with no indistinguishable
views - by Hinterstoisser et al. (ACCV'12).
:param R_est: 3x3 ndarray with the estimated rotation matrix.
:param t_est: 3x1 ndarray with the estimated translation vector.
:param R_gt: 3x3 ndarray with the ground-truth rotation matrix.
:param t_gt: 3x1 ndarray with the ground-truth translation vector.
:param pts: nx3 ndarray with 3D model points.
:return: The calculated error.
"""
pts_est = misc.transform_pts_Rt(pts, R_est, t_est)
pts_gt = misc.transform_pts_Rt(pts, R_gt, t_gt)
e = np.linalg.norm(pts_est - pts_gt, axis=1).mean()
return e
def add(pose_est, pose_gt, model):
"""
Average Distance of Model Points for objects with no indistinguishable views
- by Hinterstoisser et al. (ACCV 2012).
:param pose_est: Estimated pose given by a dictionary:
{'R': 3x3 rotation matrix, 't': 3x1 translation vector}.
:param pose_gt: The ground truth pose given by a dictionary (as pose_est).
:param model: Object model given by a dictionary where item 'pts'
is nx3 ndarray with 3D model points.
:return: Error of pose_est w.r.t. pose_gt.
"""
pts_gt = misc.transform_pts_Rt(model['pts'], pose_gt['R'], pose_gt['t'])
pts_est = misc.transform_pts_Rt(model['pts'], pose_est['R'], pose_est['t'])
e = np.linalg.norm(pts_est - pts_gt, axis=1).mean()
return e
def adi(R_est, t_est, R_gt, t_gt, model):
"""
Average Distance of Model Points for objects with indistinguishable views
- by Hinterstoisser et al. (ACCV 2012).
:param R_est, t_est: Estimated pose (3x3 rot. matrix and 3x1 trans. vector).
:param R_gt, t_gt: GT pose (3x3 rot. matrix and 3x1 trans. vector).
:param model: Object model given by a dictionary where item 'pts'
is nx3 ndarray with 3D model points.
:return: Error of pose_est w.r.t. pose_gt.
"""
pts_est = misc.transform_pts_Rt(model['pts'], R_est, t_est)
pts_gt = misc.transform_pts_Rt(model['pts'], R_gt, t_gt)
# Calculate distances to the nearest neighbors from pts_gt to pts_est
nn_index = spatial.cKDTree(pts_est)
nn_dists, _ = nn_index.query(pts_gt, k=1)
e = nn_dists.mean()
return e
def adi(R_est, t_est, R_gt, t_gt, pts):
"""Average Distance of Model Points for objects with indistinguishable views
- by Hinterstoisser et al. (ACCV'12).
:param R_est: 3x3 ndarray with the estimated rotation matrix.
:param t_est: 3x1 ndarray with the estimated translation vector.
:param R_gt: 3x3 ndarray with the ground-truth rotation matrix.
:param t_gt: 3x1 ndarray with the ground-truth translation vector.
:param pts: nx3 ndarray with 3D model points.
:return: The calculated error.
"""
pts_est = misc.transform_pts_Rt(pts, R_est, t_est)
pts_gt = misc.transform_pts_Rt(pts, R_gt, t_gt)
# Calculate distances to the nearest neighbors from vertices in the
# ground-truth pose to vertices in the estimated pose.
nn_index = spatial.cKDTree(pts_est)
nn_dists, _ = nn_index.query(pts_gt, k=1)
e = nn_dists.mean()
return e
def adi(pose_est, pose_gt, model):
"""
Average Distance of Model Points for objects with indistinguishable views
- by Hinterstoisser et al. (ACCV 2012).
:param pose_est: Estimated pose given by a dictionary:
{'R': 3x3 rotation matrix, 't': 3x1 translation vector}.
:param pose_gt: The ground truth pose given by a dictionary (as pose_est).
:param model: Object model given by a dictionary where item 'pts'
is nx3 ndarray with 3D model points.
:return: Error of pose_est w.r.t. pose_gt.
"""
pts_gt = misc.transform_pts_Rt(model['pts'], pose_gt['R'], pose_gt['t'])
pts_est = misc.transform_pts_Rt(model['pts'], pose_est['R'], pose_est['t'])
# Calculate distances to the nearest neighbors from pts_gt to pts_est
nn_index = spatial.cKDTree(pts_est)
nn_dists, _ = nn_index.query(pts_gt, k=1)
e = nn_dists.mean()
return e
def mcpd(poses_est, poses_gt, model):
"""
Maximum Corresponding Point Distance.
:param poses_est: List of poses that are indistinguishable from the
estimated pose. Each pose is given by a dictionary:
{'R': 3x3 rotation matrix, 't': 3x1 translation vector}.
:param poses_gt: List of poses that are indistinguishable from the
ground truth pose (format as for poses_est).
:param model: Object model given by a dictionary where item 'pts'
is nx3 ndarray with 3D model points.
:return: Error of pose_est w.r.t. pose_gt.
"""
e = -1
for pose_est in poses_est:
pts_est = misc.transform_pts_Rt(model['pts'], pose_est['R'], pose_est['t'])
for pose_gt in poses_gt:
pts_gt = misc.transform_pts_Rt(model['pts'], pose_gt['R'], pose_gt['t'])
dist = np.linalg.norm(pts_gt - pts_est, axis=1).max()
if dist < e or e == -1:
e = dist
return e
def mssd(R_est, t_est, R_gt, t_gt, pts, syms):
"""Maximum Symmetry-Aware Surface Distance (MSSD).
See: http://bop.felk.cvut.cz/challenges/bop-challenge-2019/
:param R_est: 3x3 ndarray with the estimated rotation matrix.
:param t_est: 3x1 ndarray with the estimated translation vector.
:param R_gt: 3x3 ndarray with the ground-truth rotation matrix.
:param t_gt: 3x1 ndarray with the ground-truth translation vector.
:param pts: nx3 ndarray with 3D model points.
:param syms: Set of symmetry transformations, each given by a dictionary with:
- 'R': 3x3 ndarray with the rotation matrix.
- 't': 3x1 ndarray with the translation vector.
:return: The calculated error.
"""
pts_est = misc.transform_pts_Rt(pts, R_est, t_est)
es = []
for sym in syms:
R_gt_sym = R_gt.dot(sym['R'])
t_gt_sym = R_gt.dot(sym['t']) + t_gt
pts_gt_sym = misc.transform_pts_Rt(pts, R_gt_sym, t_gt_sym)
es.append(np.linalg.norm(pts_est - pts_gt_sym, axis=1).max())
return min(es)