-
Notifications
You must be signed in to change notification settings - Fork 4
/
optics.py
244 lines (234 loc) · 11.6 KB
/
optics.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
import sys
import torch
import torch.nn as nn
import torch.optim as optim
import os
from odak.learn.wave import wavenumber,generate_complex_field,calculate_amplitude,propagate_beam,linear_grating,calculate_phase
from odak.learn.tools import zero_pad,crop_center,save_image
from odak.tools import check_directory
from odak import np
from data import DatasetFromFolder,load
from tqdm import tqdm
sys.path.append('../loss_functions')
def prepare(settings,wavelength):
torch.cuda.empty_cache()
torch.manual_seed(0)
cuda = settings["general"]["cuda"]
resolution = settings["slm"]["resolution"]
device = torch.device("cuda" if cuda else "cpu")
if cuda:
torch.cuda.empty_cache()
torch.random.seed()
kernel = torch.rand(
1,
2,
resolution[0],
resolution[1],
).detach().to(device).requires_grad_()
dataset = DatasetFromFolder(
settings["dataset"]["input directory"],
settings["dataset"]["output directory"],
device
)
target = load(settings["general"]["target filename"],device)
criterion = [
nn.MSELoss().to(device),
]
return kernel,target,dataset,criterion,device
def evaluate(image,target,criterion,w=[1.,]):
loss = w[0]*criterion[0](image,target)
return loss
def optimize(settings,wavelength,kernel,target,criterion,device,multiplier=1.0):
image_location = settings["image"]["location"][2]
pixel_pitch = settings["slm"]["pixel pitch"]
resolution = settings["slm"]["resolution"]
propagation_type = settings["general"]["propagation type"]
loss_weights = settings["general"]["loss weights"]
learning_rate = settings["general"]["learning rate"]
n_iterations = settings["general"]["iterations"]
m = settings["general"]["region of interest"]
ones = torch.ones(resolution[0],resolution[1],requires_grad=False).to(device)
input_phase = torch.rand(resolution[0],resolution[1]).detach().to(device).requires_grad_()
optimizer = optim.Adam([{'params': [input_phase]}],lr=learning_rate)
t = tqdm(range(n_iterations),leave=False)
mask = torch.zeros(resolution[0],resolution[1],requires_grad=False).to(device)
mask[
int(resolution[0]*m[0]):int(resolution[0]*m[1]),
int(resolution[1]*m[0]):int(resolution[1]*m[1])
] = 1
if type(kernel) != type(None):
kernel.requires_grad = False
for n in t:
optimizer.zero_grad()
field = a_single_step(
ones,
input_phase,
kernel,
image_location,
wavelength,
pixel_pitch,
propagation_type
)
image = calculate_amplitude(field)**2
loss = evaluate(image*mask,target*mask*multiplier,criterion,w=loss_weights)
description = "Iteration:{}, Loss:{:.4f}".format(n,loss.item())
loss.backward(retain_graph=True)
optimizer.step()
t.set_description(description)
if 'description' in locals():
print(description)
torch.cuda.empty_cache()
return input_phase.detach(),image.detach()
def find_kernel(settings,wavelength,kernel,dataset,criterion,device):
image_location = settings["image"]["location"][2]
pixel_pitch = settings["slm"]["pixel pitch"]
resolution = settings["slm"]["resolution"]
loss_weights = settings["general"]["loss weights"]
propagation_type = settings["general"]["propagation type"]
m = settings["kernel"]["region of interest"]
learning_rate = settings["kernel"]["learning rate"]
n_iterations = settings["kernel"]["iterations"]
optimizer = optim.Adam([{'params': [kernel]}],lr=learning_rate)
ones = torch.ones(resolution[0],resolution[1],requires_grad=False).to(device)
mask = torch.zeros(resolution[0],resolution[1],requires_grad=False).to(device)
mask[
int(resolution[0]*m[0]):int(resolution[0]*m[1]),
int(resolution[1]*m[0]):int(resolution[1]*m[1])
] = 1
t = tqdm(range(n_iterations),leave=False)
for n in t:
total_loss = 0
id_set = range(dataset.__len__())
t0 = tqdm(range(dataset.__len__()),leave=False)
for i in t0:
optimizer.zero_grad()
input_phase,target = dataset.__getitem__(i)
input_phase = input_phase*2*np.pi
field = a_single_step(
ones,
input_phase,
kernel,
image_location,
wavelength,
pixel_pitch,
propagation_type
)
image = calculate_amplitude(field)**2
loss = evaluate(image*mask,target*mask,criterion)
total_loss += loss.item()
loss.backward(retain_graph=True)
optimizer.step()
description = "Image:{}, Loss:{:.4f}".format(i,loss.item())
t0.set_description(description)
if n == n_iterations-1:
if i == 0:
save_multiple(n,i,input_phase,target,image,settings)
total_loss /= dataset.__len__()
description = "Iteration:{}, Loss:{:.4f}".format(n,total_loss)
t.set_description(description)
if 'description' in locals():
print(description)
torch.cuda.empty_cache()
return kernel
def save_multiple(n,i,input_phase,target,image,settings):
save(
input_phase,
'input_phase_{}_{}.png'.format(n,i),
directory=settings["general"]["output directory"],
save_type='phase'
)
save(
target*255.,
'output_{}_{}.png'.format(n,i),
directory=settings["general"]["output directory"],
save_type='image'
)
save(
image*255.,
'reconstruction_{}_{}.png'.format(n,i),
directory=settings["general"]["output directory"],
save_type='image'
)
def a_single_step(hologram_amplitude,hologram_phase,kernel,distance,wavelength,pixel_pitch,propagation_type):
field = generate_complex_field(hologram_amplitude,hologram_phase)
k = wavenumber(wavelength)
field_padded = zero_pad(field)
if type(kernel) == type(None):
final_field_padded = propagate_beam(field_padded,k,distance,pixel_pitch,wavelength,propagation_type)
final_field = crop_center(final_field_padded)
return final_field
final_field = torch.zeros((kernel.shape[0],kernel.shape[2],kernel.shape[3]),dtype=torch.complex64).to(field.device)
h_a = kernel[0,0]
h_p = kernel[0,1]
h = generate_complex_field(h_a,h_p)
h = zero_pad(h)
new_field_padded = propagate_beam(field_padded,None,None,None,None,propagation_type='custom',kernel=h)
new_field = crop_center(new_field_padded)
return_field = new_field
return return_field
def single_propagation(field,H):
U1 = torch.fft.fftshift(torch.fft.fft2(torch.fft.fftshift(field)))
U2 = H*U1
result = torch.fft.ifftshift(torch.fft.ifft2(torch.fft.ifftshift(U2)))
return result
def save(field,filename='output.png',directory='./',save_type='image'):
check_directory(directory)
fn = '{}/{}'.format(directory,filename)
if save_type == 'image':
field_save = (field-field.min())/(field.max()-field.min())*255.
elif save_type == 'phase':
field_save = field%(2*np.pi)/(2*np.pi)*255.
save_image(fn,field_save)
def start(settings):
wavelength = settings["beam"]["wavelength"]
directory = settings["general"]["output directory"]
multiplier = settings["kernel"]["multiplier"]
fn_kernel = './calibrations/kernel.pt'
kernel,target,dataset,criterion,device = prepare(settings,wavelength)
check_directory('./calibrations')
if os.path.isfile(fn_kernel) == True:
kernel = torch.load(fn_kernel).to(device)
settings["kernel"]["iterations"] = 0
kernel = find_kernel(
settings,
wavelength,
kernel,
dataset,
criterion,
device,
)
torch.save(kernel,fn_kernel)
checker_complex = linear_grating(
settings["slm"]["resolution"][0],
settings["slm"]["resolution"][1],
axis='y'
).to(device)
checker = calculate_phase(checker_complex)
hologram_phase,reconstruction = optimize(
settings,
wavelength,
kernel,
target,
criterion,
device,
multiplier
)
save(reconstruction,filename='reconstruction.png',directory=directory,save_type='image')
save(hologram_phase,filename='hologram_phase.png',directory=directory,save_type='phase')
save(hologram_phase+checker,filename='hologram_phase_checker.png',directory=directory,save_type='phase')
save(target,filename='target.png',directory=directory,save_type='image')
multiplier = settings["ideal"]["multiplier"]
hologram_phase_ideal,reconstruction_ideal = optimize(
settings,
wavelength,
None,
target,
criterion,
device,
multiplier
)
save(reconstruction_ideal,filename='reconstruction_ideal.png',directory=directory,save_type='image')
save(hologram_phase_ideal,filename='hologram_phase_ideal.png',directory=directory,save_type='phase')
save(hologram_phase_ideal+checker,filename='hologram_phase_ideal_checker.png',directory=directory,save_type='phase')
save(target,filename='target.png',directory=directory,save_type='image')
return True