# Choose labels
ix = rnd.choice(np.arange(Yk.shape[0]), size=1, p=Wk / np.sum(Wk))
# One-hot label image
y[Yk[ix, 0], Yk[ix, 1], k] = True
z[k, :] = np.array([Yk[ix, 0], Yk[ix, 1], k])
if vis:
fn_segs = fn + 'TRUE_scan_p' + str(n + 1) + '_r' + str(r +
1) + '.png'
plot_scan(X[:, :, 0], savefn=fn_segs)
fn_segs = fn + 'TRUE_segs_p' + str(n + 1) + '_r' + str(r +
1) + '.png'
plot_segmentation(Y, savefn=fn_segs)
# Shape of X
H, W, D = X.shape
'''Scikit's VB GMM'''
# # Initialize model
# model = BayesianGaussianMixture(n_components=K,
# max_iter=max_iter,
# verbose=3)
# # Fit model
# model.fit(X.reshape((-1, 1)))
# # Segment image
# Y_h = model.predict(X.reshape((-1, 1))).reshape((H, W))
# Choose labels
ix = rnd.choice(np.arange(Yk.shape[0]), size=1, p=Wk / np.sum(Wk))
# One-hot label image
y[Yk[ix, 0], Yk[ix, 1], k] = True
z[k, :] = np.array([Yk[ix, 0], Yk[ix, 1], k])
if vis:
fn_segs = fn + 'TRUE_scan_p' + str(n + 1) + '_r' + str(r +
1) + '.png'
plot_scan(X[30:-30, 30:-30, 0], savefn=fn_segs)
fn_segs = fn + 'TRUE_segs_p' + str(n + 1) + '_r' + str(r +
1) + '.png'
plot_segmentation(Y[30:-30, 30:-30], savefn=fn_segs)
'''Scikit's VB GMM'''
# Initialize model
model = BayesianGaussianMixture(n_components=K,
max_iter=max_iter,
verbose=3)
# Fit model
model.fit(X.reshape((-1, 1)))
# Segment image
Y_h = model.predict(X.reshape((-1, 1))).reshape((H, W))
# Obtain posteriors
post = model.predict_proba(X.reshape((-1, 1))).reshape((H, W, K))