-
Notifications
You must be signed in to change notification settings - Fork 0
/
prep_data_images.py
154 lines (116 loc) · 5.62 KB
/
prep_data_images.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
import numpy as np
import torch
import matplotlib.pyplot as plt
import subprocess
import os
from torch.utils.data import DataLoader
from sklearn.model_selection import train_test_split
from torch.nn.utils.rnn import pack_sequence, pad_packed_sequence
from PIL import Image, ImageDraw
from tqdm.auto import tqdm
def main():
dataset = ExtexifyDataset()
size = 32
os.mkdir(f"images{size}/")
for i in tqdm(range(len(dataset))):
image, y = create_ith_image_from_dataset(dataset, size, i)
if not os.path.exists(f"images{size}/{y.item()}"):
os.mkdir(f"images{size}/{y.item()}")
image.save(f"images{size}/{y.item()}/{i}.png")
print("Created dataset")
def create_ith_image_from_dataset(dataset, size, i):
stroke, y = dataset[i]
image = create_image(size, stroke)
return image, y
def create_image(size, stroke):
image = Image.new("L", (size, size), color = 0)
draw = ImageDraw.Draw(image)
for j in range(len(stroke) - 1):
if stroke[j, 2] != 1:
p1 = (stroke[j, :2].numpy() * size).tolist()
p2 = (stroke[j + 1, :2].numpy() * size).tolist()
draw.line(p1 + p2, fill=255, width=1)
else:
p1 = (stroke[j, :2].numpy() * size).tolist()
draw.point(p1, fill=255)
p1 = (stroke[-1, :2].numpy() * size).tolist()
draw.point(p1, fill=255)
return image
def collate(batch):
x, y = zip(*batch)
packed_x = pack_sequence(x, enforce_sorted = False)
y = torch.tensor(y)
return packed_x, y
def dataloaders(batch_size):
download_dataset()
dataset = ExtexifyDataset("./dataX.npy", "./dataY.npy")
dataset_train, dataset_val, dataset_test = \
dataset.split_train_val_test(0.7, 0.2, 0.1)
dataloader_train = DataLoader(dataset_train, batch_size = batch_size,\
shuffle = True, collate_fn = collate,
num_workers = os.cpu_count())
dataloader_val = DataLoader(dataset_val, batch_size = batch_size,\
shuffle = False, collate_fn = collate)
dataloader_test = DataLoader(dataset_test, batch_size = batch_size,\
shuffle = False, collate_fn = collate)
return dataloader_train, dataloader_val, dataloader_test
class ExtexifyDataset(torch.utils.data.Dataset):
def __init__(self, strokes_file = "data_processed/dataX.npy",\
labels_file = "data_processed/dataY.npy",\
strokes_labels = None):
if strokes_labels is not None:
self.strokes, self.labels = strokes_labels
else:
strokes = np.load(strokes_file, allow_pickle = True)
labels = np.load(labels_file, allow_pickle = True)
print("Loaded file")
assert all([i.shape[1] == 4 for s in strokes for i in s])
self.strokes = [torch.tensor(np.delete(np.vstack(inst), 2, axis = 1)).float() \
for inst in tqdm(strokes)]
for i, stroke in tqdm(enumerate(self.strokes), total = len(self.strokes)):
stroke[:, :2] -= stroke[:, :2].min(dim = 0).values
stroke[:, :2] /= (stroke[:, :2].max(dim = 0).values + 1e-15)
stroke[:, :2] *= 0.95
stroke[:, :2] += 0.025
print("Processed strokes")
labels_dict = {l: i for i, l in enumerate(sorted(set(labels)))}
self.labels = torch.tensor([labels_dict[i] for i in labels])
print("Processed labels")
assert len(self.strokes) == len(self.labels)
def __len__(self):
return len(self.strokes)
def __getitem__(self, idx):
return self.strokes[idx], self.labels[idx]
def split_train_val_test(self, percent_train, percent_val, percent_test):
total = percent_train + percent_val + percent_test
assert abs(total - 1) < 1e-9, total
assert 0 < percent_train and 0 < percent_val and 0 < percent_test
train, rest = self.split(percent_train)
val, test = rest.split(percent_val / (percent_test + percent_val))
return train, val, test
def split(self, percent_train):
x_train, x_test, y_train, y_test = \
train_test_split(self.strokes, self.labels, \
train_size = percent_train, random_state = 13)
train = ExtexifyDataset(strokes_labels = [x_train, y_train])
test = ExtexifyDataset(strokes_labels = [x_test, y_test])
return train, test
def download(filename, drivelink):
if not os.path.exists(filename):
confirm_cmd = ["wget", "--quiet", "--save-cookies", "/tmp/cookies.txt", \
"--keep-session-cookies", "--no-check-certificate", \
f'https://docs.google.com/uc?export=download&id={drivelink}', '-O-']
confirm = subprocess.Popen(confirm_cmd, stdout = subprocess.PIPE)
sed_cmd = ["sed", "-r", "-n", 's/.*confirm=([0-9A-Za-z_]+).*/\\1\\n/p']
sed = subprocess.Popen(sed_cmd, stdin = confirm.stdout, stdout = subprocess.PIPE)
confirm.stdout.close() # Allow p1 to receive a SIGPIPE if p2 exits.
output = sed.communicate()[0].decode('utf-8').replace("\n", "")
link = f"https://docs.google.com/uc?export=download&confirm={output}&id={drivelink}"
cmd = ["wget", "--load-cookies", "/tmp/cookies.txt", link, "-O", filename]
subprocess.run(cmd)
subprocess.run(["rm", "-rf", "/tmp/cookies.txt"])
def download_dataset():
download("./dataX.npy", "18KMxHJujq8Nb3SIMMFPHvTvQ3oBJtqXZ")
download("./dataY.npy", "1sArcVn6WCftYdtRmziy8t7V6D8Dr3Vhd")
if __name__ == "__main__":
main()