-
Notifications
You must be signed in to change notification settings - Fork 2
/
generate_test_data_linear_array_USED.py
263 lines (221 loc) · 14.5 KB
/
generate_test_data_linear_array_USED.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
#from numba import jit, cuda
import numpy as np
import os
import matplotlib.pyplot as plt
import sfs
import scipy
import tqdm
import argparse
import tensorflow as tf
from data_lib import params_linear_2D
#import jax.numpy as jnp
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
#os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'True'
print("Running test set generation")
# /nas/home/lcomanducci/soundfield_synthesis_filterbank/MeshRIR/S32-M441_npy
# /nas/home/lcomanducci/soundfield_synthesis_filterbank/MeshRIR/irutilities.py
from skimage.metrics import structural_similarity as ssim
#from scikit-image.metrics import structural_similarity as _ssim
#plt.rcParams.update({
#"text.usetex": True,
#"font.family": "sans-serif",
#"font.sans-serif": ["Helvetica"],
#'font.size': 20})
#@jit(target_backend='cuda')
def plot_soundfield(cmap, P, n_f, selection, axis_label_size, tick_font_size, save_path, plot_ldspks=True, do_norm=True):
figure = plt.figure(figsize=(20, 20))
if do_norm:
im = sfs.plot2d.amplitude(np.reshape(P[:, n_f], (params_linear_2D.N_sample, params_linear_2D.N_sample)),
params_linear_2D.grid, xnorm=[0, 0, 0], cmap=cmap, vmin=-1.0, vmax=1.0, colorbar=False)
else:
im = sfs.plot2d.amplitude(np.reshape(P[:, n_f], (params_linear_2D.N_sample, params_linear_2D.N_sample)),
params_linear_2D.grid, cmap=cmap, colorbar=False, vmin=P[:, n_f].min(), vmax=P[:, n_f].max(), xnorm=None)
if plot_ldspks:
sfs.plot2d.loudspeakers(params_linear_2D.array.x[selection], params_linear_2D.array.n[selection], a0=1, size=0.18)
plt.xlabel('$x [m]$', fontsize=axis_label_size), plt.ylabel('$y [m]$', fontsize=axis_label_size)
plt.tick_params(axis='both', which='major', labelsize=tick_font_size)
cbar = plt.colorbar(im, fraction=0.046)
cbar.ax.tick_params(labelsize=tick_font_size)
# cbar.set_label('$NRE~[\mathrm{dB}]$',fontsize=tick_font_size))
plt.tight_layout()
plt.savefig(save_path)
plt.show()
#@jit(target_backend='cuda')
def sound_field(d, selection, secondary_source, array, grid, tapering=True):
if tapering:
tapering_window = sfs.tapering.tukey(selection, alpha=0.3)
else:
tapering_window = sfs.tapering.none(selection)
p = sfs.fd.synthesize(d, tapering_window, array, secondary_source, grid=grid)
return p, tapering_window
def normalize(x):
min_x = x.min()
max_x = x.max()
#print("x = {}\nmin_x = {}\nmax_x = {}".format(x, min_x, max_x))
den = (max_x-min_x)
if den == 0 :
den = den + np.finfo(np.float128).eps # 1.084202172485504434e-19
x_norm = (x - min_x)/den
return x_norm
def nmse(P_hat, P_gt,type='freq'):
if type=='freq':
return np.mean((np.power(np.abs(P_hat[ :, :] - P_gt[ :, :]), 2) / np.power(np.abs(P_gt[ :, :]), 2)), axis=0)
else:
return np.power(np.abs(P_hat[ :, :] - P_gt[ :, :]), 2) / np.power(np.abs(P_gt[ :, :]), 2)
def ssim_abs(P_hat, P_gt):
P_hat = normalize(np.abs(P_hat))
P_gt = normalize(np.abs(P_gt))
return ssim(P_gt, P_hat, data_range=1)
def ssim_freq(P_hat, P_gt):
ssim_freq_array = np.zeros(params_linear_2D.N_freqs)
for n_f in range(params_linear_2D.N_freqs):
ssim_freq_array[n_f] = ssim_abs(P_hat[:, n_f], P_gt[:, n_f])
return ssim_freq_array
#@jit(target_backend='cuda')
def main():
# Arguments parse
parser = argparse.ArgumentParser(description='Generate data for linear array setup')
parser.add_argument('--dataset_path', type=str, help="Base Data Directory", default='/nas/home/ralessandri/thesis_project/dataset/test/linear_array')
parser.add_argument('--model_path', type=str, help='Deep learning models folder', default='/nas/home/ralessandri/thesis_project/models/linear_array')
#parser.add_argument('--dataset_path', type=str, help="Base Data Directory", default='C:/Users/rales/OneDrive/Desktop/POLIMI/TESI/pressure_matching_deep_learning/dataset/test/linear_array')
#parser.add_argument('--model_path', type=str, help='Deep learning models folder', default='C:/Users/rales/OneDrive/Desktop/POLIMI/TESI/pressure_matching_deep_learning/models/linear_array')
parser.add_argument('--gt_soundfield', type=bool, help='compute ground truth soundfield', default=True)
#parser.add_argument('--wfs', type=bool, help='compute soundfield synthesis', default=True)
#parser.add_argument('--hoa', type=bool, help='compute higher order ambisonics', default=True)
parser.add_argument('--pm', type=bool, help='compute pressure matching', default=True)
parser.add_argument('--pwd_cnn', type=bool, help='compute model-based acoustic rendering + CNN', default=False) # True?
parser.add_argument('--n_loudspeakers', type=int, help='Numbr of loudspeakers in array', default=16)
parser.add_argument('--n_missing', type=int, help='number missing loudspeakers', default=0)
eval_points = True
PLOT = False
args = parser.parse_args()
dataset_path =args.dataset_path
# Grid of points where we actually compute the soundfield
point = params_linear_2D.point
#N_pts = len(point) #params_circular.radius_sources_test
N_pts = int((len(params_linear_2D._grid_[0][0])))
grid = params_linear_2D._grid_
# Load green function secondary sources --> eval points (it is in train directory since it is the same)
dataset_path_train = '/nas/home/ralessandri/thesis_project/dataset/linear_array'
#dataset_path_train = 'C:/Users/rales/OneDrive/Desktop/POLIMI/TESI/pressure_matching_deep_learning/dataset/linear_array'
#green_function_sec_sources_path = 'green_function_sec_sources_nl_' + str(args.n_loudspeakers) + '_r_' + str(params_linear_2D.rangeX[0]) + '.npy'
green_function_sec_sources_path = 'green_function_sec_sources_nl_' + str(params_linear_2D.N_lspks) + '_r_' + str(params_linear_2D.rangeX[0]) + '.npy'
g_path_joined = os.path.join(dataset_path_train, green_function_sec_sources_path).replace('\\','/')
print("problem = ", g_path_joined)
#G = np.load(os.path.join(dataset_path_train, green_function_sec_sources_path))
G = np.load(g_path_joined)
# Load Missing loudspeakers configuration
lspk_config_path = 'lspk_config_nl_' + str(params_linear_2D.N_lspks) + '_missing_' + str(args.n_missing) + '.npy'
lspk_config_path_global = os.path.join(dataset_path_train, 'setup', lspk_config_path)
# Let's precompute what we need in order to apply the selected models
# Load pwd_cnn deep learning model
if args.pwd_cnn:
model_name = 'model_linear_config_nl_'+str(args.n_loudspeakers)
network_model = tf.keras.models.load_model(os.path.join(args.model_path, model_name).replace('\\', '/'))
if args.pm:
lambda_ = 1e-2
G_cp = G[params_linear_2D.idx_lr[params_linear_2D.idx_cp]] # Green function at control points
# points_cp = params_circular.point[params_circular.idx_lr[params_circular.idx_cp]]
C_pm = np.zeros((args.n_loudspeakers, len(params_linear_2D.idx_cp), params_linear_2D.N_freqs ),dtype=complex)
for n_f in tqdm.tqdm(range(len(params_linear_2D.wc))):
# INSTALLARE JAXLIB
C_pm[:, :, n_f] = np.matmul(np.linalg.pinv(np.matmul(G_cp[:, :, n_f].transpose(), G_cp[:, :, n_f]) + lambda_ * np.eye(args.n_loudspeakers)),G_cp[:, :, n_f].transpose())
#C_pm[:, :, n_f] = np.matmul(jnp.linalg.pinv(np.matmul(G_cp[:, :, n_f].transpose(), G_cp[:, :, n_f]) + lambda_ * np.eye(args.n_loudspeakers)),G_cp[:, :, n_f].transpose())
#if eval_points:
#N_pts = len(params_linear_2D.idx_lr)
#G = G[params_linear_2D.idx_lr]
#point = params_linear_2D.point_lr
nmse_pwd_cnn = np.zeros((len(params_linear_2D.src_pos_train), params_linear_2D.num_sources, params_linear_2D.N_freqs))
nmse_pwd_pm = np.zeros_like(nmse_pwd_cnn)
ssim_pwd_cnn = np.zeros_like(nmse_pwd_cnn)
ssim_pwd_pm = np.zeros_like(nmse_pwd_cnn)
print("n_r lenght = ", len(params_linear_2D.src_pos_train))
print("n_s length = ", params_linear_2D.num_sources)
for n_r in tqdm.tqdm(range(len(params_linear_2D.src_pos_train))):
P_gt = np.zeros((N_pts, params_linear_2D.N_freqs), dtype=complex)
P_pwd_cnn = np.zeros((N_pts, params_linear_2D.N_freqs), dtype=complex)
P_pwd_pm = np.zeros((N_pts, params_linear_2D.N_freqs), dtype=complex)
for n_s in range(params_linear_2D.num_sources):
if PLOT:
n_s = 41
if (n_r < len(params_linear_2D.src_pos_test[0]) & n_s < len(params_linear_2D.src_pos_test[1])):
#print("src_pos_test[0 -> {}, 1 -> {}]".format(len(params_linear_2D.src_pos_test[0]), len(params_linear_2D.src_pos_test[1])))
xs = params_linear_2D.src_pos_test[n_r, n_s]
if args.gt_soundfield:
for n_f in range(params_linear_2D.N_freqs):
hankel_arg = (params_linear_2D.wc[n_f] / params_linear_2D.c) * np.linalg.norm(point[:, :2] - xs, axis=1)
P_gt[:, n_f] = (1j / 4) * scipy.special.hankel2(0, hankel_arg)
if args.pwd_cnn:
if eval_points:
P_input = P_gt[params_linear_2D.idx_cp]
else:
P_input = P_gt[params_linear_2D.idx_lr[params_linear_2D.idx_cp]]
d_array_cnn = network_model.predict(
np.expand_dims(np.concatenate([np.real(P_input), np.imag(P_input)], axis=0), axis=[0, -1]).astype('float32'))[0, :, :, 0].astype('float64')
d_array_cnn_complex = d_array_cnn[:int(d_array_cnn.shape[0] / 2)] + (1j * d_array_cnn[int(d_array_cnn.shape[0] / 2):])
plt.figure(figsize=(10,10))
plt.imshow(d_array_cnn.T, aspect='auto',cmap='RdBu')
plt.xlabel('$l$',fontsize=120),plt.ylabel('$k$',fontsize=120)
plt.tick_params(axis='both', which='major', labelsize=120)
plt.gca().invert_yaxis()
plt.show()
for n_p in range(N_pts):
P_pwd_cnn[ n_p, :] = np.sum(G[n_p] * d_array_cnn_complex, axis=0)
if args.pm:
d_pm = np.zeros((args.n_loudspeakers, params_linear_2D.N_freqs),dtype=complex)
for n_f in range(params_linear_2D.N_freqs):
if eval_points:
d_pm[:, n_f] = np.matmul(C_pm[:, :, n_f], P_gt[params_linear_2D.idx_cp, n_f]) # CHECK idx_cp is USING EVAL POINTSSSSS
else:
d_pm[:, n_f] = np.matmul(C_pm[:, :, n_f], P_gt[params_linear_2D.idx_lr[params_linear_2D.idx_cp], n_f]) # CHECK idx_cp is USING EVAL POINTSSSSS
for n_p in range(N_pts):
P_pwd_pm[n_p, :] = np.sum(G[n_p] * d_pm, axis=0)
#nmse_pwd_cnn[n_r, n_s], nmse_pwd_pm[n_r, n_s] = #nmse(P_pwd_cnn, P_gt), nmse(P_pwd_pm, P_gt)
#ssim_pwd_cnn[n_r, n_s], ssim_pwd_pm[n_r, n_s] = ssim_freq(P_pwd_cnn, P_gt), ssim_freq(P_pwd_pm, P_gt)
if PLOT:
plt.figure(), plt.plot(params_linear_2D.array_pos[:4, 0], params_linear_2D.array_pos[:4, 1], 'r*'), plt.show()
d = np.linalg.norm(np.array([params_linear_2D.array_pos[1, 0], params_linear_2D.array_pos[1, 1]])-np.array([params_linear_2D.array_pos[2, 0], params_linear_2D.array_pos[2, 1]]))
aliasing_freq = params_linear_2D.c/(2*d)
# Plot params
selection = np.ones_like(params_linear_2D.array_pos[:, 0])
selection = selection == 1
n_f = 63 # 63
print(str(params_linear_2D.f_axis[n_f]))
cmap = 'RdBu_r'
tick_font_size = 70
axis_label_size = 90
# Ground truth
plot_paths = os.path.join('plots', 'linear')
save_path = os.path.join(plot_paths, 'sf_real_source_' + str(n_s) + '_f_' + str(params_linear_2D.f_axis[n_f]) + '_nl'+str(args.n_loudspeakers)+'.pdf')
plot_soundfield(cmap, P_gt, n_f, selection, axis_label_size, tick_font_size, save_path, plot_ldspks=False)
# PWD-CNN
save_path = os.path.join(plot_paths, 'sf_pwd_cnn_' + str(n_s) + '_f_' + str(params_linear_2D.f_axis[n_f]) + '_nl'+str(args.n_loudspeakers)+'.pdf')
plot_soundfield(cmap, P_pwd_cnn, n_f, selection, axis_label_size, tick_font_size, save_path)
# Error
#nmse_pwd_cnn = 10*np.log10(nmse(P_pwd_cnn, P_gt, type='full'))
#save_path = os.path.join(plot_paths, 'nmse_pwd_cnn_' + str(n_s) + '_f_' + str(params_linear_2D.f_axis[n_f]) + '_nl'+str(args.n_loudspeakers)+'.pdf')
plot_soundfield(cmap, nmse_pwd_cnn, n_f, selection, axis_label_size, tick_font_size, save_path, do_norm=False)
# PM
save_path = os.path.join(plot_paths, 'sf_pm_' + str(n_s) + '_f_' + str(params_linear_2D.f_axis[n_f]) + '_nl'+str(args.n_loudspeakers)+'.pdf')
plot_soundfield(cmap, P_pwd_pm, n_f, selection, axis_label_size, tick_font_size, save_path)
# Error
#nmse_pm = 10*np.log10(nmse(P_pwd_pm, P_gt, type='full'))
#save_path = os.path.join(plot_paths, 'nmse_pm_' + str(n_s) + '_f_' + str(params_linear_2D.f_axis[n_f]) +'_nl'+str(args.n_loudspeakers)+ '.pdf')
#plot_soundfield(cmap, nmse_pm, n_f, selection, axis_label_size, tick_font_size, save_path, do_norm=False)
plt.figure()
#plt.plot(params_linear_2D.f_axis, 10*np.log10(np.mean(nmse(P_pwd_cnn, P_gt, type='full'), axis=0)),'k-*')
#plt.plot(params_linear_2D.f_axis, 10*np.log10(np.mean(nmse(P_pwd_pm, P_gt, type='full'), axis=0)),'r-*')
plt.tick_params(axis='both', which='major', labelsize=10)
plt.legend(['CNN', 'PM'])
plt.show()
print('pause')
# If we are plotting it means we are just computing data for the paper --> no need to save anything
if not PLOT:
# Save arrays
np.savez(os.path.join(dataset_path, 'nmse_nl_' + str(args.n_loudspeakers)+'.npz'), nmse_pwd_cnn=nmse_pwd_cnn, nmse_pwd_pm=nmse_pwd_pm)
np.savez(os.path.join(dataset_path, 'ssim_nl_' + str(args.n_loudspeakers)+'.npz'), ssim_pwd_cnn=ssim_pwd_cnn, ssim_pwd_pm=ssim_pwd_pm)
if __name__ == '__main__':
main()
print("Ended Generate linear testing data")