rMSBP-python/reconstruct_sim_mbs.py at master · bu-cisl/rMSBP-python · GitHub

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
import logging
import os
import json
import numpy as np
import init_helper

import ssnp
from pycuda import gpuarray
from tools import rev_ng_acc
from ssnp.utils.fista import get_q
from tv2d_temp import prox_tv_Michael

logging.info('version 1115')

base_dir = "2dres_mbs"

sim_name = 'h06_scale04'
measurements = np.sqrt(ssnp.read(os.path.join(base_dir, f"sim/{sim_name}.tiff"), gpu=False) / 0.4)

measurements = gpuarray.to_gpu(measurements)

with open(os.path.join(base_dir, "exp_bfdf.json")) as f:
    metadata = json.load(f)

parameters = {
    # data parameters
    "slices":     8,
    "z_step":     1,
    "xy_patch":   (1024, 1024),
    "overlap":    0,

    # reconstruction parameters
    "iterations": 50,
    "gamma":      5,
    "tau":        8e-7,
    "model_type": "bpm"
}

illumination = metadata['illumination'][:]
LEDs = len(illumination)
assert LEDs == len(measurements)
N0 = metadata['n_medium']

positive = False
iterations = parameters['iterations']
xy_patch = parameters['xy_patch']
z_step = parameters['z_step']
SLICES = parameters['slices']
GAMMA = parameters['gamma']
TAU = parameters['tau']
u_in = ssnp.read("plane", np.complex128, shape=(LEDs, *xy_patch), gpu=False)
n = np.zeros((SLICES, *xy_patch), dtype=np.double)
n = gpuarray.to_gpu(n)
ng = n.copy()
x_1 = n.copy()

beam = ssnp.BeamArray(u_in[0])
beam.config.lambda0 = metadata['lambda']
beam.config.xyz = (*metadata['xy_size'], z_step)
beam.config.n0 = N0
pupil = beam.multiplier.binary_pupil(metadata['NA'] / N0, gpu=True)
for i in range(LEDs):
    c_ab = [c / N0 for c in illumination[i]]
    u_in[i] = beam.multiplier.tilt(c_ab, trunc=True, gpu=False)

illumination = np.array(illumination)

ng_acc = rev_ng_acc(ng)

for step in range(iterations):
    logging.info(f"{step=} started")

    loss_tot = 0
    for ill_i in range(109):
        ng.fill(0)
        beam.forward = u_in[ill_i]
        with beam.track():
            for i in range(4):
                beam.bpm(0.5, n[i])
                beam.bpm(10)
            # beam.conj()
            beam *= -1
            for i in range(4, 8):
                beam.bpm(10)
                beam.bpm(0.5, n[i])
            beam.bpm(-42)
            beam.a_mul(pupil)
            loss = beam.forward_mse_loss(measurements[ill_i]) * LEDs
            logging.debug(f"{ill_i = }, {loss = :.9f}")
            loss_tot += loss
        beam.tape.collect_gradient({'n': ng_acc}, reverse=True)
        # tape_grad = beam.tape.collect_gradient({'n': ng_acc, 'u1_in': None}, reverse=True)
        # u_in_grad_arr = tape_grad['u1_in'][0]
        # beam.recycle_array(u_in_grad_arr)
        # u_in_grad_arr = u_in_grad_arr.get()
        # u_in_grad = np.real(np.sum(u_in_grad_arr * np.conj(u_in[ill_i])))
        #
        # # Nesterov
        # norm_x[ill_i] = normalization[ill_i] - u_in_grad * (0.3 if ill_i < 25 else 5)
        # normalization[ill_i] = norm_x[ill_i] + (norm_x[ill_i] - norm_x1[ill_i]) * (get_q(step) - 1) / get_q(step + 1)
        # norm_x1[ill_i] = norm_x[ill_i]

        assert next(ng_acc) is None
        step_size = parameters['gamma']

        ng *= step_size
        n -= ng
    x = ng  # only memory assignment
    if positive:
        prox_fgp(n, GAMMA * TAU, out=x)
    else:
        prox_tv_Michael(n, GAMMA * TAU, out=x)

    n.set(x)
    # x.set(n)

    n -= x_1
    n *= (get_q(step) - 1) / get_q(step + 1)
    logging.info((get_q(step) - 1) / get_q(step + 1))
    n += x
    x_1.set(x)
    logging.info(f"Step {step} finished, {loss_tot = :.9f}")


tiff_range = (-0.08, 0.08)
ssnp.write(os.path.join(base_dir, f"rec/h10_BFDF_reg_{N0 + tiff_range[0]:.3f}_{N0 + tiff_range[1]:.3f}.tiff"), n,
           pre_operator=lambda x: (x - tiff_range[0]) / (tiff_range[1] - tiff_range[0]))
# np.save(os.path.join(base_dir, f"norm_hist.npy"), norm_history)
# ssnp.write(os.path.join(base_dir, "rec/5l_45l.npy"), n)