def train(cls, X, frac, kmax):
n_samples = X.shape[1]
n_points = X.shape[0] // 2
# align shapes
Y = procrustes(X)
# compute rigid transform
R = calc_rigid_basis(Y)
# project out rigidity
P = R.T.dot(Y)
dY = Y - R.dot(P)
# compute non-rigid transformation
D = pca(dY, frac, min(kmax, n_samples - 1, n_points - 1))
k = D.shape[1]
# combine subspaces
V = np.concatenate((R, D), axis=1)
# project raw data onto subspace
Q = V.T.dot(X)
# normalize coordinates w.r.t. scale
for i in xrange(n_samples):
Q[:, i] /= Q[0, i]
# compute variance
e = np.empty(4 + k, dtype=float)
Q = pow(Q, 2)
e[:4] = -1 # no clamping for rigid body coefficients
e[4:] = Q[4:].sum(axis=1) / (n_samples - 1)
# return model
return cls(V, e)
def train(cls, lmks, imgs, ref, frac, kmax):
smodel = ShapeModel.load('data/shape.npz')
tmodel = TextureModel.load('data/texture.npz')
# build diagonal matrix of weights that measures
# the unit difference between shape and texture parameters
r = tmodel.variance.sum() / smodel.variance.sum()
Ws = r * np.eye(smodel.num_modes())
n_samples = lmks.shape[1]
tm = TextureMapper(480, 640)
# create empty matrix
n_feats = smodel.num_modes() + tmodel.num_modes()
A = np.empty((n_feats, n_samples))
# concatenate shape and texture parameters for each training example
for i in xrange(n_samples):
img = next(imgs)
lmk = lmks[:,i]
sparams = smodel.calc_params(lmk)
tparams = tmodel.calc_params(img, lmk, ref, tm.warp_triangles)
# ignore first 4 shape parameters
A[:,i] = np.concatenate((Ws.dot(sparams[4:]), tparams))
D = pca(A, frac, kmax)
# compute variance
Q = pow(D.T.dot(A), 2)
e = Q.sum(axis=1) / (n_samples-1)
return cls(D, e, Ws)
def train(cls, lmks, imgs, ref, frac, kmax):
smodel = ShapeModel.load('data/shape.npz')
tmodel = TextureModel.load('data/texture.npz')
# build diagonal matrix of weights that measures
# the unit difference between shape and texture parameters
r = tmodel.variance.sum() / smodel.variance.sum()
Ws = r * np.eye(smodel.num_modes())
n_samples = lmks.shape[1]
tm = TextureMapper(480, 640)
# create empty matrix
n_feats = smodel.num_modes() + tmodel.num_modes()
A = np.empty((n_feats, n_samples))
# concatenate shape and texture parameters for each training example
for i in xrange(n_samples):
img = next(imgs)
lmk = lmks[:, i]
sparams = smodel.calc_params(lmk)
tparams = tmodel.calc_params(img, lmk, ref, tm.warp_triangles)
# ignore first 4 shape parameters
A[:, i] = np.concatenate((Ws.dot(sparams[4:]), tparams))
D = pca(A, frac, kmax)
# compute variance
Q = pow(D.T.dot(A), 2)
e = Q.sum(axis=1) / (n_samples - 1)
return cls(D, e, Ws)
def train(cls, X, frac, kmax):
n_samples = X.shape[1]
n_points = X.shape[0] // 2
# align shapes
Y = procrustes(X)
# compute rigid transform
R = calc_rigid_basis(Y)
# project out rigidity
P = R.T.dot(Y)
dY = Y - R.dot(P)
# compute non-rigid transformation
D = pca(dY, frac, min(kmax, n_samples-1, n_points-1))
k = D.shape[1]
# combine subspaces
V = np.concatenate((R,D), axis=1)
# project raw data onto subspace
Q = V.T.dot(X)
# normalize coordinates w.r.t. scale
for i in xrange(n_samples):
Q[:,i] /= Q[0,i]
# compute variance
e = np.empty(4+k, dtype=float)
Q = pow(Q, 2)
e[:4] = -1 # no clamping for rigid body coefficients
e[4:] = Q[4:].sum(axis=1) / (n_samples-1)
# return model
return cls(V, e)
def train(cls, lmks, imgs, ref, frac, kmax):
G = get_data_matrix(imgs, lmks, ref)
Gm = G.mean(axis=1)
G -= Gm[:,np.newaxis]
N = lmks.shape[1]
D = pca(G, frac, kmax)
# normalize eigenvectors
for i in range(D.shape[1]):
D[:,i] /= np.linalg.norm(D[:,i])
# compute variance
Q = D.T.dot(G)
Q = pow(Q, 2)
e = Q.sum(axis=1) / (N-1)
return cls(D, Gm, e)
def train(cls, lmks, imgs, ref, frac, kmax):
G = get_data_matrix(imgs, lmks, ref)
Gm = G.mean(axis=1)
G -= Gm[:,np.newaxis]
N = lmks.shape[1]
D = pca(G, frac, kmax)
# normalize eigenvectors
for i in xrange(D.shape[1]):
D[:,i] /= np.linalg.norm(D[:,i])
# compute variance
Q = D.T.dot(G)
Q = pow(Q, 2)
e = Q.sum(axis=1) / (N-1)
return cls(D, Gm, e)