forked from pidahbus/deep-image-orientation-angle-detection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgenerator.py
226 lines (174 loc) · 6.92 KB
/
generator.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
from utils import rotate_preserve_size
import glob
import os
import numpy as np
import cv2
import random
import tensorflow as tf
import os
import pandas as pd
from tensorflow.keras.utils import Sequence
from transformers import ViTFeatureExtractor
feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224')
class ViTRotGenerator(Sequence):
def __init__(self, image_dir, batch_size, dim):
self.files = glob.glob(os.path.join(image_dir, "*.jpg"))
self.batch_size = batch_size
self.dim = dim
def __len__(self):
if len(self.files) % self.batch_size == 0:
return len(self.files) // self.batch_size
return len(self.files) // self.batch_size + 1
def __getitem__(self, idx):
batch_slice = slice(idx * self.batch_size, (idx + 1) * self.batch_size)
batch_files = self.files[batch_slice]
X_conv = []
X_vit = []
y = []
for i, f in enumerate(batch_files):
try:
angle = float(np.random.choice(range(0, 360)))
img = rotate_preserve_size(f, angle, (self.dim, self.dim))
img = np.array(img)
X_vit.append(img)
img = np.expand_dims(img, axis=0)
X_conv.append(img)
y.append(angle)
except:
pass
X_vit = feature_extractor(images=X_vit, return_tensors="pt")["pixel_values"]
X_vit = np.array(X_vit)
X_conv = np.concatenate(X_conv, axis=0)
y = np.array(y)
return [X_vit, X_conv], y
def on_epoch_end(self):
random.shuffle(self.files)
class ViTValidationTestGenerator(Sequence):
def __init__(self, image_dir, df_label_path, batch_size, dim, mode):
self.image_dir = image_dir
self.batch_size = batch_size
self.dim = dim
self.mode = mode
df_label = pd.read_csv(df_label_path)
self.df = df_label[df_label["mode"] == self.mode].reset_index(drop=True)
def __len__(self):
total = self.df.shape[0]
if total % self.batch_size == 0:
return total // self.batch_size
return total // self.batch_size + 1
def __getitem__(self, idx):
batch_slice = slice(idx * self.batch_size, (idx + 1) * self.batch_size)
df_batch = self.df[batch_slice].reset_index(drop=True).copy()
X_conv = []
X_vit = []
y = []
for i in range(len(df_batch)):
try:
angle = df_batch.angle[i]
path = os.path.join(self.image_dir, df_batch.image[i])
img = rotate_preserve_size(path, angle, (self.dim, self.dim))
img = np.array(img)
X_vit.append(img)
img = np.expand_dims(img, axis=0)
X_conv.append(img)
y.append(angle)
except:
pass
X_vit = feature_extractor(images=X_vit, return_tensors="pt")["pixel_values"]
X_vit = np.array(X_vit)
X_conv = np.concatenate(X_conv, axis=0)
y = np.array(y)
return [X_vit, X_conv], y
class RotGenerator(Sequence):
def __init__(self, image_dir, batch_size, dim, channels_first=False, is_vit=False):
self.files = glob.glob(os.path.join(image_dir, "*.jpg"))
self.batch_size = batch_size
self.dim = dim
self.channels_first = channels_first
self.is_vit = is_vit
def __len__(self):
if len(self.files) % self.batch_size == 0:
return len(self.files) // self.batch_size
return len(self.files) // self.batch_size + 1
def __getitem__(self, idx):
batch_slice = slice(idx * self.batch_size, (idx + 1) * self.batch_size)
batch_files = self.files[batch_slice]
# X = np.zeros(shape=(len(batch_files), self.dim, self.dim, 3))
# y = np.zeros(shape=(len(batch_files), ))
X = []
y = []
for i, f in enumerate(batch_files):
try:
angle = float(np.random.choice(range(0, 360)))
img = rotate_preserve_size(f, angle, (self.dim, self.dim))
img = np.array(img)
if self.is_vit:
X.append(img)
else:
if self.channels_first:
img = img.transpose(2, 0, 1)
img = np.expand_dims(img, axis=0)
X.append(img)
# X[i] = img
# y[i] = angle
y.append(angle)
except:
pass
if self.is_vit:
X = feature_extractor(images=X, return_tensors="pt")["pixel_values"]
X = np.array(X)
else:
X = np.concatenate(X, axis=0)
y = np.array(y)
return X, y
def on_epoch_end(self):
random.shuffle(self.files)
# In[83]:
class ValidationTestGenerator(Sequence):
def __init__(self, image_dir, df_label_path, batch_size, dim, mode, channels_first=False, is_vit=False):
self.image_dir = image_dir
self.batch_size = batch_size
self.dim = dim
self.mode = mode
self.channels_first = channels_first
self.is_vit = is_vit
df_label = pd.read_csv(df_label_path)
df_label["angle"] = df_label["angle"].astype("float")
self.df = df_label[df_label["mode"] == self.mode].reset_index(drop=True)
def __len__(self):
total = self.df.shape[0]
if total % self.batch_size == 0:
return total // self.batch_size
return total // self.batch_size + 1
def __getitem__(self, idx):
batch_slice = slice(idx * self.batch_size, (idx + 1) * self.batch_size)
df_batch = self.df[batch_slice].reset_index(drop=True).copy()
# X = np.zeros(shape=(len(df_batch), self.dim, self.dim, 3))
# y = np.zeros(shape=(len(df_batch), ))
X = []
y = []
for i in range(len(df_batch)):
try:
angle = df_batch.angle[i]
path = os.path.join(self.image_dir, df_batch.image[i])
img = rotate_preserve_size(path, angle, (self.dim, self.dim))
img = np.array(img)
if self.is_vit:
X.append(img)
else:
if self.channels_first:
img = img.transpose(2, 0, 1)
img = np.expand_dims(img, axis=0)
X.append(img)
# X[i] = img
# y[i] = angle
y.append(angle)
except:
pass
if self.is_vit:
X = feature_extractor(images=X, return_tensors="pt")["pixel_values"]
X = np.array(X)
else:
X = np.concatenate(X, axis=0)
y = np.array(y)
return X, y