obj_id, 'step')
elif p['error_type'] == 'mssd':
if not spheres_overlap:
e = [float('inf')]
else:
e = [
pose_error.mssd(R_e, t_e, R_g, t_g,
models[obj_id]['pts'],
models_sym[obj_id])
]
elif p['error_type'] == 'mspd':
e = [
pose_error.mspd(R_e, t_e, R_g, t_g, K,
models[obj_id]['pts'],
models_sym[obj_id])
]
elif p['error_type'] in ['ad', 'add', 'adi']:
if not spheres_overlap:
# Infinite error if the bounding spheres do not overlap. With
# typically used values of the correctness threshold for the AD
# error (e.g. k*diameter, where k = 0.1), such pose estimates
# would be considered incorrect anyway.
e = [float('inf')]
else:
if p['error_type'] == 'ad':
if obj_id in dp_model['symmetric_obj_ids']:
e = [
pose_error.adi(
def evaluate_sequence(renderer, obj_id, Re, te, Rview, tview, K, pts, Rgt, tgt, syms, depth_test, Rs, ts):
"""
Evaluate target object under all given pose estimates [Re, te]. Values for MSSD, MSPD, VSD and ours are computed.
"""
# BOP toolkit settings and parameters
delta, taus, normalized_by_diameter, diameter, r2 = 15, [0.2], True, models_info[obj_id]['diameter'],\
models_meta[obj_id]['frobenius_scale']
im_width = 640
theta_adi, theta_mssd, theta_mspd, theta_vsd = 0.1, 0.2, 20, 0.3
metric, phi = 'frobenius', 20
# ground-truth pose in camera space
Rgt_cam = Rview @ Rgt
tgt_cam = Rview @ tgt + tview
Rps = []
tps = []
errors = []
for R, t in zip(Re, te):
# pose in camera space
Rcam = Rview @ R
tcam = Rview @ t + tview
# reshape estimate for BOP toolkit
tgt = tgt.reshape(3, 1)
tgt_cam = tgt_cam.reshape(3, 1)
t = t.reshape(3, 1)
tcam = tcam.reshape(3, 1)
# compute baseline pose-error functions
mssd = error.mssd(R, t, Rgt, tgt, pts, syms) / diameter
mspd = error.mspd(Rcam, tcam, Rgt_cam, tgt_cam, K, pts, syms) * (640/im_width)
vsd = np.mean(error.vsd(Rcam, tcam, Rgt_cam, tgt_cam, depth_test, K, delta, taus,
normalized_by_diameter, diameter, renderer, obj_id))
# ours: match estimate to closest plausible pose
Rp, tp, dp = plausibility.find_closest(R, t, Rs, ts, metric, pts, syms, r2)
# matched plausible pose in camera space; reshape for BOP toolkit
Rp_cam = Rview @ Rp
tp_cam = (Rview @ tp + tview).reshape(3, 1)
tp = tp.reshape(3, 1)
# ours: compute implausibility and physical plausibility error terms
implausibility = np.clip(dp / phi, 0, 1)
mssd_pp = mssd + theta_mssd * implausibility
mspd_pp = mspd + theta_mspd * implausibility
vsd_pp = vsd + theta_vsd * implausibility
mssd_op = mssd + error.mssd(R, t, Rp, tp, pts, syms) / diameter
mspd_op = mspd + error.mspd(Rcam, tcam, Rp_cam, tp_cam, K, pts, syms) * (640/im_width)
vsd_op = vsd + np.mean(error.vsd(Rcam, tcam, Rp_cam, tp_cam, depth_test, K, delta, taus,
normalized_by_diameter, diameter, renderer, obj_id))
errors.append([mssd, mspd, vsd, implausibility,
mssd_pp, mspd_pp, vsd_pp,
mssd_op, mspd_op, vsd_op])
Rps.append(Rp)
tps.append(tp)
return np.array(errors), np.array(Rps).reshape(-1, 3, 3), np.array(tps).reshape(-1, 3)
