def main():
print(f'Loading MP data from {MP_FPATH}...')
mp_data = loadmat(MP_FPATH)
pts = mp_data['means']
wts = mp_data['weights']
cov = mp_data['cov']
print('Computing transformations...')
alphas = []
betas = []
for t in range(N_FRAMES - 1):
print(f'frames: {t} -> {t+1}')
pts_1 = pts[t, 0:N_MPS, :]
wts_1 = wts[t, 0:N_MPS, 0]
pts_2 = pts[t + 1, 0:N_MPS, :]
wts_2 = wts[t + 1, 0:N_MPS, 0]
alpha, beta, _ = imagereg.ot_reg_linear_1(pts_1, pts_2, wts_1, wts_2)
alphas.append(alpha)
betas.append(beta)
print(f'Saving results to {OUT_FPATH}')
mat_dict = {
'fname': MP_FPATH,
'n_frames': N_FRAMES,
'n_mps': N_MPS,
'alphas': np.array(alphas),
'betas': np.array(betas),
}
savemat(OUT_FPATH, mat_dict)
def main():
print(f'Loading MP data from {MP_FPATH}...')
mp_data = loadmat(MP_FPATH)
pts = mp_data['means']
wts = mp_data['weights']
cov = mp_data['cov']
print('Computing transformations...')
# Points and weights for first frame
pts_0 = pts[0, 0:N_MPS, :]
wts_0 = wts[0, 0:N_MPS, 0]
f_alpha = [None] * N_FRAMES
f_beta = [None] * N_FRAMES
rec_pts = [None] * N_FRAMES
f_alpha[0] = np.array([0, 0, 0])
f_beta[0] = np.eye(3)
rec_pts[0] = pts_0
for t in range(1, N_FRAMES):
print(f'frame: {t}')
# Points and weight distribution for previous frame
pts_1 = pts[t - 1, 0:N_MPS, :]
wts_1 = wts[t - 1, 0:N_MPS, 0]
# Points and weight distribution for current frame
pts_2 = pts[t, 0:N_MPS, :]
wts_2 = wts[t, 0:N_MPS, 0]
# Compute mapping from reconstruction of prev. frame to current frame
alpha, beta, _ = ot_reg_linear_1(pts_1, pts_2, wts_1, wts_2)
print('det(beta)')
print(np.linalg.det(beta))
# Compute mapping from first frame to current frame using recursive update
f_alpha[t] = beta @ f_alpha[t - 1] + alpha
f_beta[t] = beta @ f_beta[t - 1]
# Use mapping to reconstruct current frame from first frame
rec_pts[t] = f_alpha[t] + pts_0 @ f_beta[t].T
print(f'Saving results to {OUT_FPATH}')
mat_dict = {
'fname': MP_FPATH,
'n_frames': N_FRAMES,
'n_mps': N_MPS,
'f_alpha': np.array(f_alpha),
'f_beta': np.array(f_beta),
'rec_pts': np.array(rec_pts),
}
savemat(OUT_FPATH, mat_dict)
def main():
print(f'Loading MP data from {MP_FPATH}...')
mp_data = loadmat(MP_FPATH)
pts = mp_data['means'][1:N_FRAMES + 1, :, :]
wts = mp_data['weights'][1:N_FRAMES + 1, :, :]
cov = mp_data['cov']
print('Computing transformations...')
# Points and weights for first frame
pts_0 = pts[1, 0:N_MPS, :]
wts_0 = wts[1, 0:N_MPS, 0]
wts_unif = np.ones(N_MPS) / N_MPS
f_alpha = [None] * N_FRAMES
f_beta = [None] * N_FRAMES
p_mtx = [None] * N_FRAMES
f_alpha[0] = np.array([0, 0, 0])
f_beta[0] = np.eye(3)
p_mtx[0] = np.eye(N_MPS)
for t in range(1, N_FRAMES):
print(f'frame: {t}')
pts_t = pts[t, 0:N_MPS, :]
wts_t = wts[t, 0:N_MPS, 0]
alpha, beta, log = ot_reg_linear_1(pts_0, pts_t, wts_unif, wts_unif)
f_alpha[t] = alpha
f_beta[t] = beta
p_mtx[t] = log['P']
print(f'Saving results to {OUT_FPATH}')
mat_dict = {
'fname': MP_FPATH,
'n_frames': N_FRAMES,
'n_mps': N_MPS,
'f_alpha': np.array(f_alpha),
'f_beta': np.array(f_beta),
'p_mtx': np.array(p_mtx),
}
savemat(OUT_FPATH, mat_dict)