-
Notifications
You must be signed in to change notification settings - Fork 0
/
fltlnd_imports.py
354 lines (261 loc) · 12.6 KB
/
fltlnd_imports.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
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
from sklearn.model_selection import train_test_split
import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import map_coordinates
from keras.utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, BatchNormalization
from keras.optimizers import Adam
from keras.callbacks import ReduceLROnPlateau
from keras.metrics import CategoricalAccuracy
def train_val_test_split(X, y, random_state = 42):
X_train, X1, y_train, y1 = train_test_split(X, y, test_size=0.4, random_state = random_state)
X_val, X_test, y_val, y_test = train_test_split(X1, y1, test_size=0.5, random_state = random_state)
return(X_train, X_val, X_test, y_train, y_val, y_test)
#----SHEAR----
def random_shear_image(input_image, shear_range=(-0.2, 0.2)):
input_image = 255*input_image
# Get the dimensions of the input image
image_height, image_width = input_image.shape
# Generate random shear factors for x and y directions
shear_x = np.random.uniform(shear_range[0], shear_range[1])
shear_y = np.random.uniform(shear_range[0], shear_range[1])
# Create a grid of coordinates for the original image
x, y = np.meshgrid(np.arange(image_width), np.arange(image_height))
# Apply shear transformation to the coordinates
sheared_x = x + shear_x * y
sheared_y = y + shear_y * x
# Initialize the sheared image
sheared_image = np.zeros((image_height, image_width), dtype=np.uint8)
# Interpolate pixel values from the original image to sheared coordinates
sheared_image = np.zeros_like(input_image)
# Use scipy's map_coordinates for efficient interpolation
map_coordinates(input_image, (sheared_y, sheared_x), output=sheared_image, order=1, mode='nearest')
return (sheared_image/255)
def augment_dataset_with_shear(images, y_train, proportion = 0.2, shear_range=(-0.2, 0.2)):
augmented_images = []
for image in images:
augmented_image = random_shear_image(image, shear_range)
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----GAUSSIAN BLUR----
def add_gaussian_blur(image, kernel_size=5, sigma=1.0):
image = image*255
# Create a Gaussian kernel
kernel = np.fromfunction(
lambda x, y: (1/(2*np.pi*sigma**2)) * np.exp(-((x - (kernel_size-1)/2)**2 + (y - (kernel_size-1)/2)**2) / (2*sigma**2)),
(kernel_size, kernel_size)
)
kernel /= np.sum(kernel) # Normalize the kernel
# Apply convolution using FFT (Fast Fourier Transform)
blurred_image = np.fft.ifft2(np.fft.fft2(image) * np.fft.fft2(kernel, s=image.shape))
return (np.real(blurred_image)/255)
def augment_dataset_with_blur(images, y_train, proportion = 0.2, kernel_size = 5, sigma = 1.0):
augmented_images = []
for image in images:
augmented_image = add_gaussian_blur(image, kernel_size, sigma)
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----NOISE----
def add_gaussian_noise(image, mean=0, std_dev=25):
noise = np.random.normal(mean, std_dev, image.shape)
noisy_image = image*255 + noise
noisy_image = np.clip(noisy_image, 0, 255) # Clip values to the valid grayscale range (0-255)
noisy_image = noisy_image.astype(np.uint8)/255 # Convert the result back to an 8-bit unsigned integer
return noisy_image
def augment_dataset_with_noise(images, y_train, proportion = 0.2, mean = 0, std_dev=25):
augmented_images = []
for image in images:
augmented_image = add_gaussian_noise(image)
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----INVERTION----
def invert(image):
adjusted_image = 255 - image*255
return (np.clip(adjusted_image, 0, 255).astype(np.uint8)/255)
def augment_dataset_with_invertion(images, y_train, proportion = 0.2, min_brightness=80, max_brightness=100):
augmented_images = []
for image in images:
augmented_image = invert(image)
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----CONTRAST----
def lower_contrast(image, contrast_factor=0.5):
adjusted_image = contrast_factor * (image*255 - 128) + 128
return (np.clip(adjusted_image, 0, 255).astype(np.uint8)/255)
def augment_dataset_with_contrast(images, y_train, proportion=0.2, contrast_factor=0.5):
augmented_images = []
for image in images:
augmented_image = lower_contrast(image, contrast_factor * np.random.uniform(0.6, 1.4))
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----BRIGHTNESSS----
def adjust_brightness(image, brightness_factor):
adjusted_image = image*255 + brightness_factor
return (np.clip(adjusted_image, 0, 255).astype(np.uint8)/255)
def augment_dataset_with_brightness(images, y_train, proportion = 0.2, min_brightness=80, max_brightness=100):
augmented_images = []
for image in images:
# Randomly choose a brightness adjustment factor within the specified range
brightness_factor = np.random.uniform(min_brightness, max_brightness)
if (np.random.uniform(0, 1) < 0.5):
brightness_factor *= -1
augmented_image = adjust_brightness(image, brightness_factor)
augmented_images.append(augmented_image)
augmented_dataset = np.array(augmented_images)
X_aug, _, y_aug, _ = train_test_split(augmented_dataset, y_train, test_size=1-proportion, random_state=42)
return(X_aug,y_aug)
#----MERGE AND SHUFFLE----
def merge_and_shuffle(X_train, y_train, X_aug, y_aug):
# Assuming X1_train, X2_train, y1_train, and y2_train are numpy arrays
# If not, you can convert them using np.array(your_list)
# Reshape the image data to be 1D arrays
X1_train_flat = X_train.reshape(-1, 50 * 50)
X2_train_flat = X_aug.reshape(-1, 50 * 50)
# Merge the datasets horizontally (along columns)
X_combined = np.vstack((X1_train_flat, X2_train_flat))
y_combined = np.hstack((y_train, y_aug))
# Combine X_combined and y_combined
combined_data = np.column_stack((X_combined, y_combined))
# Shuffle the combined dataset
np.random.shuffle(combined_data)
# Separate the shuffled data back into X and y
X_shuffled = combined_data[:, :-1].reshape(-1, 50, 50) # Reshape back to 3D image data
y_shuffled = combined_data[:, -1]
# Now, X_shuffled and y_shuffled contain your merged and shuffled image data.
return(X_shuffled, y_shuffled)
def merge_and_shuffle_for_one_hot(X_train, y_train, X_aug, y_aug):
# Assuming X1_train, X2_train, y1_train, and y2_train are numpy arrays
# If not, you can convert them using np.array(your_list)
# Reshape the image data to be 1D arrays
X1_train_flat = X_train.reshape(-1, 50 * 50)
X2_train_flat = X_aug.reshape(-1, 50 * 50)
# Merge the datasets horizontally (along columns)
X_combined = np.vstack((X1_train_flat, X2_train_flat))
y_combined = np.vstack((y_train, y_aug))
# Combine X_combined and y_combined
combined_data = np.column_stack((X_combined, y_combined))
# Shuffle the combined dataset
np.random.shuffle(combined_data)
print(combined_data[:,:-5].shape)
X_shuffled = combined_data[:, :-5].reshape(-1, 50, 50) # Reshape back to 3D image data
y_shuffled = combined_data[:, -5:]
# Now, X_shuffled and y_shuffled contain your merged and shuffled image data.
return X_shuffled, y_shuffled
#----PLOTTING----
def plot_image(image, label):
"""
Plots a grayscale image with a size of 50x50.
parameters: i - i-th image
"""
plt.imshow(image, cmap='gray',vmin=0, vmax=255)
plt.axis('off')
plt.title(label)
plt.show()
# Define your custom plot_image() function
def plot_images(image, label, ax):
ax.imshow(image, cmap='gray',vmin=0, vmax=1)
ax.axis('off')
ax.set_title(str(label))
def plot10(X, y, offset=0):
# Create a 2x5 grid of subplots
fig, axs = plt.subplots(2, 5, figsize=(12, 6))
for i in range(2):
for j in range(5):
index = i * 5 + j
if index < 10:
plot_images(X[index+offset], y[index+offset], axs[i, j])
# Adjust subplot spacing and display the figure
plt.tight_layout()
plt.show()
#----AUGMENTATION----
def augment(X_train, y_train, datagen):
# Reshape the image to (1, height, width, channels) for ImageDataGenerator
X_train_reshaped = X_train.reshape((X_train.shape[0], 50, 50, 1))
# Generate augmented images
augmented_images = datagen.flow(X_train_reshaped, y_train)
# Initialize lists for X and y batches
X_batch_list = []
y_batch_list = []
# Collect batches into X and y lists
for _ in range(len(augmented_images)):
X_batch, y_batch = augmented_images.next()
X_batch_list.append(X_batch)
y_batch_list.append(y_batch)
# Concatenate X and y batches into single NumPy arrays
X_batches = np.concatenate(X_batch_list, axis=0)
y_batches = np.concatenate(y_batch_list, axis=0)
X_aug = X_batches.reshape(X_train.shape[0], 50, 50)
y_aug = y_batches
return (X_aug, y_aug)
#----MIXUP----
def mixup_data(X, y, mixup_ratio, size=0.2):
assert len(X) == len(y), "X and y must have the same length"
# Create a dictionary that groups indices by their corresponding labels
label_indices = {}
for idx, label in enumerate(y):
label = tuple(label) # Convert the label to a hashable tuple
if label not in label_indices:
label_indices[label] = []
label_indices[label].append(idx)
X_mixed = []
y_mixed = []
for _ in range(int(size * len(X))):
# Randomly select two different label groups
l1, l2 = np.random.choice((0,1,2,3,4), size=2, replace=False)
label_group_1 = np.zeros(5, dtype=int)
label_group_2 = np.zeros(5, dtype=int)
label_group_1[l1] = 1
label_group_2[l2] = 1
# Randomly select indices from the two selected label groups
i = np.random.choice(label_indices[tuple(label_group_1)])
j = np.random.choice(label_indices[tuple(label_group_2)])
# Randomly sample the mixup ratio λ from a Beta distribution
lambda_value = np.random.beta(mixup_ratio, mixup_ratio)
# Perform mixup for the features
mixed_features = lambda_value * X[i] + (1 - lambda_value) * X[j]
X_mixed.append(mixed_features)
# Perform mixup for the labels (assuming one-hot encoding)
mixed_labels = lambda_value * y[i] + (1 - lambda_value) * y[j]
y_mixed.append(mixed_labels)
return np.array(X_mixed), np.array(y_mixed)
#----TRAIN AND EVAL----
def train_and_eval_pipe(model, X_train, y_train, X_val, y_val, epochs_number = 10):
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.8, patience=10, min_lr=3e-4*(0.2**2))
#Train the model
history = model.fit(X_train, y_train, batch_size=256,
validation_data = (X_val, y_val),epochs=epochs_number,
verbose = 0, callbacks=[reduce_lr])
# Print the Train and val accuracies
m = CategoricalAccuracy()
y_pred_train = model.predict(X_train, verbose = 0)
m.update_state(y_train,y_pred_train)
print("Train set accuracy:", m.result().numpy())
m.reset_state()
y_pred_val = model.predict(X_val, verbose = 0)
m.update_state(y_val, y_pred_val)
print("Validation set accuracy:", m.result().numpy())
# Plot training history
plt.plot(history.history['loss'], label='Train Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.title('Training and Validation Loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()
def test_set_accuracy(model, X_test, y_test):
m = CategoricalAccuracy()
y_pred_test = model.predict(X_test, verbose = 0)
m.update_state(y_test, y_pred_test)
print("Test set accuracy:", m.result().numpy())