pystemsim/generate_MoS2_dataset.py at main · AustinHouston/pystemsim · 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
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
import os
import sys
import numpy as np
import secrets
from PIL import Image
from ase.io import read
sys.path.append('/Users/austin/Documents/GitHub/DataGenSTEM/DataGenSTEM')
import data_generator as dg

# Read in xtal
print('reading cif files:')
xtal_l1 = read(filename = './crystal_files/MoS2_l1.cif')
xtal_l2 = read(filename = './crystal_files/MoS2_l2.cif')
combined_xtal = xtal_l1 + xtal_l2
positions = combined_xtal.get_positions()[:, :2]
xmin, xmax = np.min(positions[:, 0]), np.max(positions[:, 0])
ymin, ymax = np.min(positions[:, 1]), np.max(positions[:, 1])
borders = 1
axis_extent = (xmin - borders, xmax + borders, ymin - borders, ymax + borders)

image_counter = 0
used_seeds = []
n_images = 5000

# Simulation based images - first 1000
crop_size = 512
pixel_size = 0.078125 # Angstrom/pixel, determines number of points, aka resolution of maps.  the xtal determines the fov
n_crops = 10 # number of crops per large image

print('making images from gb simulation:')
while image_counter < 1000:
    print(image_counter, ' out of ', n_images)

    # Set numpy randomizer
    master_seed = secrets.randbits(128)
    while master_seed in used_seeds:
        master_seed = secrets.randbits(128)
    used_seeds.append(master_seed)
    rng = np.random.default_rng(master_seed)

    # Set random params
    # --------------------------------------------------
    phonon_sigma = rng.uniform(0.02, 0.1)
    rotation_l1 = rng.uniform(0, 360)
    rotation_l2 = rotation_l1 + rng.choice([-15, 0, 15])
    atom_var = rng.normal(loc = 0.175, scale = 0.01)
    airy_disk_size = 1
    shot_noise = rng.uniform(0.8, 0.94)
    magnification_var = rng.uniform(0.2, 0.35)
    crop_param_seed = rng.integers(0, 1000000)
    # --------------------------------------------------

    # Make xtal (with random rotations and vacancies)
    rot_xtal_l1 = dg.get_xtal_matrix(xtal = xtal_l1, n_cells = (1,1,1), rotation = rotation_l1, n_vacancies = 10, phonon_sigma = phonon_sigma, axis_extent = axis_extent)
    rot_xtal_l2 = dg.get_xtal_matrix(xtal = xtal_l2, n_cells = (1,1,1), rotation = rotation_l2, n_vacancies = 10, phonon_sigma = phonon_sigma, axis_extent = axis_extent)

    # Make potential and psf (with random atom_size and psf size)
    potential_l1 = dg.get_pseudo_potential(xtal = rot_xtal_l1, pixel_size = pixel_size, sigma = atom_var, axis_extent = axis_extent)
    potential_l2 = dg.get_pseudo_potential(xtal = rot_xtal_l2, pixel_size = pixel_size, sigma = atom_var, axis_extent = axis_extent)
    potential_l1, potential_l2 = potential_l1 / np.max(potential_l1), potential_l2 / np.max(potential_l2)
    potential_total = potential_l1 + potential_l2

    # Make point spread function
    psf = dg.get_point_spread_function(airy_disk_radius = airy_disk_size, size = 32)
    psf_resize = dg.resize_image(np.array(psf), n = max(potential_total.shape)) # for plotting on same axes as image

    # Make image, masks and add shot noise
    perfect_image = dg.convolve_kernel(potential_total, psf)
    noisy_image = dg.add_poisson_noise(perfect_image, shot_noise = shot_noise)
    masks_l1 = dg.get_masks(rot_xtal_l1, axis_extent = axis_extent, pixel_size = pixel_size, radius = 5, mode='one_hot')
    masks_l2 = dg.get_masks(rot_xtal_l2, axis_extent = axis_extent, pixel_size = pixel_size, radius = 5, mode='one_hot')

    # Try only the atom masks (do not include the first, background mask)
    masks_l1 = masks_l1[1:]
    masks_l2 = masks_l2[1:]

    # Crop and zoom
    batch_ims = dg.shotgun_crop(noisy_image, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, roi = 'middle')
    batch_ims = batch_ims.reshape(-1,crop_size,crop_size)

    batch_masks_l1 = dg.shotgun_crop(masks_l1, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, return_binary = True, roi = 'middle')
    batch_masks_l2 = dg.shotgun_crop(masks_l2, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, return_binary = True, roi = 'middle')
    batch_masks = np.stack((batch_masks_l1, batch_masks_l2), axis=1)
    batch_masks = batch_masks.reshape(-1,len(masks_l1)+len(masks_l2),crop_size,crop_size)
    batch_masks = (batch_masks > 0.5).astype(int) # binarize the masks

    # save images and masks
    for image, label_set in zip(batch_ims, batch_masks):
        if np.average(image) > 0.05: # only save images with atoms - this is a hacky way to do this
            image = image - np.min(image)
            image = image / np.max(image)
            img = Image.fromarray((image * 255).astype(np.uint8))
            img.save(f'/Users/austin/Desktop/MOS2_dataset/images/image_{image_counter:04d}.png')

            os.makedirs(f'/Users/austin/Desktop/MOS2_dataset/labels/label_{image_counter:04d}/', exist_ok=True)
            for j, label in enumerate(label_set):
                img = Image.fromarray((label * 255).astype(np.uint8))
                img.save(f'/Users/austin/Desktop/MOS2_dataset/labels/label_{image_counter:04d}/class_{j:01d}.png')

            image_counter += 1


print('making images from pristine MoS2 w/ holes:')
xtal = read('./crystal_files/MoS2.cif')
# Remove bottom layer atoms (bilayer cif)
com = xtal.get_center_of_mass()
atoms_to_delete = [atom.index for atom in xtal if atom.position[2] < com[2]]
del xtal[atoms_to_delete]
xtal.center()

xtal_l1 = xtal.repeat((100,100,1))
xtal_l2 = xtal_l1.copy()
combined_xtal = xtal_l1 + xtal_l2
positions = combined_xtal.get_positions()[:, :2]
xmin, xmax = np.min(positions[:, 0]), np.max(positions[:, 0])
ymin, ymax = np.min(positions[:, 1]), np.max(positions[:, 1])
borders = 1
axis_extent = (xmin - borders, xmax + borders, ymin - borders, ymax + borders)

n_crops = 20 # number of crops per large image
while image_counter < n_images:
    print(image_counter, ' out of ', n_images)

    # Set numpy randomizer
    master_seed = secrets.randbits(128)
    while master_seed in used_seeds:
        master_seed = secrets.randbits(128)
    used_seeds.append(master_seed)
    rng = np.random.default_rng(master_seed)

    # Set random params
    # --------------------------------------------------
    phonon_sigma = rng.uniform(0.02, 0.1)
    rotation_l1 = rng.uniform(0, 360)
    rotation_l2 = rotation_l1 + rng.choice([-60, -30, -15, 0, 0, 0, 15, 30, 60])
    hole_size = rng.uniform(2, 5)
    n_holes = rng.integers(1, 10)
    atom_var = rng.normal(loc = 0.175, scale = 0.01)
    airy_disk_size = 1
    shot_noise = rng.uniform(0.8, 0.94)
    magnification_var = rng.uniform(0.2, 0.35)
    crop_param_seed = rng.integers(0, 1000000)
    # --------------------------------------------------

    # Make xtal (with random rotations and vacancies)
    rot_xtal_l1 = dg.get_xtal_matrix(xtal = xtal_l1, n_cells = (1,1,1), rotation = rotation_l1, n_vacancies = 10, phonon_sigma = phonon_sigma, axis_extent = axis_extent)
    rot_xtal_l2 = dg.get_xtal_matrix(xtal = xtal_l2, n_cells = (1,1,1), rotation = rotation_l2, n_vacancies = 10, phonon_sigma = phonon_sigma, axis_extent = axis_extent)

    # Make holes
    rot_xtal_l1 = dg.make_holes(rot_xtal_l1, n_holes = n_holes, hole_size = hole_size)
    rot_xtal_l2 = dg.make_holes(rot_xtal_l2, n_holes = n_holes, hole_size = hole_size)

    # Make potential and psf (with random atom_size and psf size)
    potential_l1 = dg.get_pseudo_potential(xtal = rot_xtal_l1, pixel_size = pixel_size, sigma = atom_var, axis_extent = axis_extent)
    potential_l2 = dg.get_pseudo_potential(xtal = rot_xtal_l2, pixel_size = pixel_size, sigma = atom_var, axis_extent = axis_extent)
    potential_l1, potential_l2 = potential_l1 / np.max(potential_l1), potential_l2 / np.max(potential_l2)
    potential_total = potential_l1 + potential_l2

    # Make point spread function
    psf = dg.get_point_spread_function(airy_disk_radius = airy_disk_size, size = 32)
    psf_resize = dg.resize_image(np.array(psf), n = max(potential_total.shape)) # for plotting on same axes as image

    # Make image, masks and add shot noise
    perfect_image = dg.convolve_kernel(potential_total, psf)
    noisy_image = dg.add_poisson_noise(perfect_image, shot_noise = shot_noise)
    masks_l1 = dg.get_masks(rot_xtal_l1, axis_extent = axis_extent, pixel_size = pixel_size, radius = 5, mode='one_hot')
    masks_l2 = dg.get_masks(rot_xtal_l2, axis_extent = axis_extent, pixel_size = pixel_size, radius = 5, mode='one_hot')

    # Try only the atom masks (do not include the first, background mask)
    masks_l1 = masks_l1[1:]
    masks_l2 = masks_l2[1:]

    # Crop and zoom
    batch_ims = dg.shotgun_crop(noisy_image, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, roi = 'middle')
    batch_ims = batch_ims.reshape(-1,crop_size,crop_size)

    batch_masks_l1 = dg.shotgun_crop(masks_l1, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, return_binary = True, roi = 'middle')
    batch_masks_l2 = dg.shotgun_crop(masks_l2, crop_size = crop_size, n_crops = n_crops, seed = crop_param_seed, magnification_var = magnification_var, return_binary = True, roi = 'middle')
    batch_masks = np.stack((batch_masks_l1, batch_masks_l2), axis=1)
    batch_masks = batch_masks.reshape(-1,len(masks_l1)+len(masks_l2),crop_size,crop_size)
    batch_masks = (batch_masks > 0.5).astype(int) # binarize the masks

    # save images and masks
    for image, label_set in zip(batch_ims, batch_masks):
        if np.average(image) > 0.05: # only save images with atoms - this is a hacky way to do this
            image = image - np.min(image)
            image = image / np.max(image)
            img = Image.fromarray((image * 255).astype(np.uint8))
            img.save(f'/Users/austin/Desktop/MOS2_dataset/images/image_{image_counter:04d}.png')

            os.makedirs(f'/Users/austin/Desktop/MOS2_dataset/labels/label_{image_counter:04d}/', exist_ok=True)
            for j, label in enumerate(label_set):
                img = Image.fromarray((label * 255).astype(np.uint8))
                img.save(f'/Users/austin/Desktop/MOS2_dataset/labels/label_{image_counter:04d}/class_{j:01d}.png')

            image_counter += 1