x, z = generate_dataset(img_shape,
n=n,
image_path=spo_dataset.__path__[0] +
'/source_image/illusion.jpg')
x_test = x[:n_train]
z_test = z[:n_train]
x_train = x[n_train:]
z_train = z[n_train:]
# ---- Initialize ---- #
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist, img_shape)
# ---- Load weights ---- #
df.load_weights('model_weights_illusion')
# ---- initialize particle filter ---- #
image_path = spo_dataset.__path__[0] + '/source_image/illusion.jpg'
ref_img = cv2.imread(image_path, 0)
ref_img = cv2.resize(ref_img, img_shape, interpolation=cv2.INTER_AREA)
# %%
pf = ParticleFilter_deep(Np=10000,
No=2,
ref_img=ref_img,
radiuses=[18, 22],
initial_pose=[[14, 18], [108, 104]],
beta=3,
likelihood=df)
n = 600
n_test = 300
n_train = n - n_test
# ---- Get the dataset ---- #
# ---- initialize particle filter ---- #
ref_img = np.array(data.checkerboard()).astype(np.float64)
ref_img = cv2.resize(ref_img, img_shape,interpolation = cv2.INTER_AREA)
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist,img_shape)
# ---- Load weights ---- #
df.load_weights('model_weights_partially_observed_chekkers')
pf = ParticleFilter_deep(Np = 100,
No = 1,
ref_img = ref_img,
radiuses = [18],
initial_pose = [[20,10]],
beta = 60,
likelihood=df)
# ---- Get the dataset ---- #
x, z = generate_dataset(img_shape = img_shape,
n = n,
image_type = "checkers",
partial=True)
n=n,
image_path=spo_dataset.__path__[0] + '/source_image/tree.jpg',
mask=spo_dataset.__path__[0] + '/source_image/tree_masked.jpg',
partial=True)
x_test = x[:n_train]
z_test = z[:n_train]
x_train = x[n_train:]
z_train = z[n_train:]
# ---- Initialize DF ---- #
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist, img_shape)
# ---- Load weights ---- #
df.load_weights('model_weights_partially_observed_tree')
# ---- Initialize testing arrays ---- #
# ---- initialize particle filter ---- #
image_path = spo_dataset.__path__[0] + '/source_image/tree.jpg'
ref_img = cv2.imread(image_path, 0)
ref_img = cv2.resize(ref_img, img_shape, interpolation=cv2.INTER_AREA)
pf = ParticleFilter_deep(Np=5000,
No=1,
ref_img=ref_img,
radiuses=[20],
initial_pose=[[14, 20]],
beta=1,
likelihood=df)
# ---- Initialize ---- #
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist, img_shape)
# ---- Train ---- #
train_likelihood(df, x_train, z_train, epochs=130) #100
train_predictor(df, x_train, epochs=10, min_img=2) #5
train_predictor(df, x_train, epochs=15, min_img=20) #10
train_update(df, x_train, z_train, epochs=10, min_img=2) #10
train_relax(df, x_train, z_train, epochs=5, min_img=10) # 10
train_relax(df, x_train, z_train, epochs=50, min_img=50) # 10
train_relax(df, x_train, z_train, epochs=50, min_img=None) # 10
# ---- Or load weights ---- #
df.load_weights('model_weights')
# ---- Test and viualize ---- #
x_old = x_test[:hist, ...].copy()
frames = []
obs_frames = []
state_frames = []
df_frames = []
direct_frames = []
for t in range(0 + hist, n_test - 1):
z_new = z_test[t].copy()
z_new_test = z_test[t].copy()
x_new = x_test[t].copy()
x_hat_df = df.predict_mean(x_old, z_new)
x_hat_df = x_hat_df[:, :, 0]
x_hat_df_like = df.estimate(z_new_test)
x, z = get_dataset_rotating_objects(image_shape=img_shape,
n=n,
var=var,
Ber=True)
x_test = np.array(x[:n_test])
z_test = np.array(z[:n_test])
x_train = np.array(x[n_test:])
z_train = np.array(z[n_test:])
# ---- Initialize DF ---- #
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist, img_shape)
# ---- Load weights ---- #
df.load_weights('model_weights_rectangles_very_noisy')
# ---- initialize particle filter ---- #
pf = ParticleFilterRect(Np=10000,
var=var,
img_shape=img_shape,
beta=1,
Ber=True)
# ---- Initialize testing arrays ---- #
cm_err_df = []
mass_err_df = []
img_err_df = []
img_kl_df = []
cm_err_pf = []
n_test = 300
n_train = n - n_test
# ---- Get the dataset ---- #
z, x = generate_dataset(img_shape, n=n, image_type="checkers")
x_test = x[:n_train]
z_test = z[:n_train]
x_train = x[n_train:]
z_train = z[n_train:]
# ---- Initialize ---- #
tf.keras.backend.clear_session()
df = DeepNoisyBayesianFilter(hist, img_shape)
# ---- Load weights ---- #
df.load_weights('model_weights_inverse_chekkers')
# ---- initialize particle filter ---- #
ref_img = np.array(data.checkerboard()).astype(np.float64)
ref_img = cv2.resize(ref_img, img_shape, interpolation=cv2.INTER_AREA)
pf = ParticleFilter_inv(Np=10000,
No=1,
ref_img=ref_img,
radiuses=[20],
initial_pose=[[14, 18]],
beta=1)
# ---- Test and viualize ---- #
x_old = x_test[:hist, ...].copy()
frames = []