-
Notifications
You must be signed in to change notification settings - Fork 0
/
train & test.py
137 lines (100 loc) · 5.55 KB
/
train & test.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
import os
import tensorflow as tf
from tensorflow import keras
import random
from keras.callbacks import TensorBoard
#from keras import backend as K
from skimage.io import imread, imshow
from skimage.transform import resize
import matplotlib.pyplot as plt
from tqdm import tqdm
import numpy as np
from PIL import Image
from cxn_model import *
IMG_WIDTH = 256
IMG_HEIGHT = 256
IMG_CHANNELS = 3
TRAIN_PATH_R = '/content/drive/My Drive/Colab Notebooks/dataset/B4/train/' #change path
TRAIN_PATH_G = '/content/drive/My Drive/Colab Notebooks/dataset/B3/train/'
TRAIN_PATH_B = '/content/drive/My Drive/Colab Notebooks/dataset/B2/train/'
TEST_PATH_R = '/content/drive/My Drive/Colab Notebooks/dataset/B4/test/'
TEST_PATH_G = '/content/drive/My Drive/Colab Notebooks/dataset/B3/test/'
TEST_PATH_B = '/content/drive/My Drive/Colab Notebooks/dataset/B2/test/'
X_train = np.zeros((350, IMG_WIDTH, IMG_WIDTH, IMG_CHANNELS), dtype=np.float32)
Y_train = np.zeros((350, IMG_WIDTH, IMG_WIDTH, 1), dtype=np.float32)
img = np.zeros((IMG_WIDTH, IMG_WIDTH, IMG_CHANNELS), dtype=np.float32)
tr=np.zeros(350)
te=np.zeros(10)
for i in range(350):
tr[i]=i;
for i in range(10):
te[i]=i;
for n, id_ in tqdm(enumerate(tr),total=350):
red = Image.open(TRAIN_PATH_R + str(int(id_)) + '.png').convert('L')
green = Image.open(TRAIN_PATH_G + str(int(id_)) + '.png').convert('L')
blue = Image.open(TRAIN_PATH_B + str(int(id_)) + '.png').convert('L')
rgb = Image.merge("RGB",(red,green,blue))
img_b = np.asarray(rgb)
#img_r = imread(TRAIN_PATH_R + str(int(id_)) + '.png')[:,:,:IMG_CHANNELS]
#img_g = imread(TRAIN_PATH_G + str(int(id_)) + '.png')[:,:,:IMG_CHANNELS]
#img_b = imread(TRAIN_PATH_B + str(int(id_)) + '.png')[:,:,:IMG_CHANNELS]
#img_r = resize(img_r, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
#img_g = resize(img_g, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
#img_b = resize(img_b, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
img_b = resize(img_b, (IMG_HEIGHT, IMG_WIDTH, 3), mode='constant', preserve_range=True)
img_b=img_b/255.0
#for i in range(256):
# img[i] = np.concatenate((img_r[i],img_g[i],img_b[i]), axis=1)
X_train[n] = img_b
#mask = Image.open('/content/drive/My Drive/Colab Notebooks/dataset/BQA/train/' + str(int(102+id_)) + '.png').convert('L')
mask = imread('/content/drive/My Drive/Colab Notebooks/dataset/BQA/train/' + str(int(id_)) + '.png')[:,:,:IMG_CHANNELS]
mask1 = resize(mask, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
#mask1 = np.asarray(mask)
#mask1 = resize(mask1, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
Y_train[n] =mask1/255.0
for io in range(256):
for jo in range(256):
if (Y_train[n][io][jo]>0.3):
Y_train[n][io][jo]=1
else:
Y_train[n][io][jo]=0
# for test images
X_test = np.zeros((10, IMG_WIDTH, IMG_WIDTH, IMG_CHANNELS), dtype=np.float32)
Y_test = np.zeros((10, IMG_WIDTH, IMG_WIDTH, 1), dtype=np.float32)
img = np.zeros((IMG_WIDTH, IMG_WIDTH, IMG_CHANNELS), dtype=np.float32)
sizes_test = []
for n, id_ in tqdm(enumerate(te),total=10):
red = Image.open(TEST_PATH_R + str(170+int(id_)) + '.png').convert('L')
green = Image.open(TEST_PATH_G + str(170+int(id_)) + '.png').convert('L')
blue = Image.open(TEST_PATH_B + str(170+int(id_)) + '.png').convert('L')
rgb = Image.merge("RGB",(red,green,blue))
img_b = np.asarray(rgb)
img_b = resize(img_b, (IMG_HEIGHT, IMG_WIDTH, 3), mode='constant', preserve_range=True)
img_b=img_b/255.0
#for i in range(256):
# img[i] = np.concatenate((img_r[i],img_g[i],img_b[i]), axis=1)
X_test[n] = img_b
mask = imread('/content/drive/My Drive/Colab Notebooks/dataset/BQA/test/' + str(int(170+id_)) + '.png')[:,:,:IMG_CHANNELS]
mask1 = resize(mask, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
#mask1 = np.asarray(mask)
#mask1 = resize(mask1, (IMG_HEIGHT, IMG_WIDTH, 1), mode='constant', preserve_range=True)
Y_test[n] =mask1/255.0
for io in range(256):
for jo in range(256):
if (Y_test[n][io][jo]>0.3):
Y_test[n][io][jo]=1
else:
Y_test[n][io][jo]=0
model = model_arch(input_rows=256, input_cols=256, num_of_channels=3, num_of_classes=1)
model.compile(optimizer = Adam(lr = 1e-4), loss = jacc_coef, metrics = [jacc_coef,'accuracy'])
#checkpointer = tf.keras.callbacks.ModelCheckpoint('model_for_nuclei.h5' , verbose=1, save_best_only=True)
#callbacks = [tf.keras.callbacks.EarlyStopping(patience=50, monitor='val_loss'),tf.keras.callbacks.TensorBoard(log_dir="logs")]
results = model.fit(X_train, Y_train, validation_split=0.05, batch_size=2, epochs=70, verbose=1) #, callbacks=[cp_callback])
preds_train = model.predict(X_train, verbose=1)
preds_val = model.predict(X_train[int(X_train.shape[0]*0.95):], verbose=1)
preds_test = model.predict(X_test, verbose=1)
preds_train_t = (preds_train > 0.5).astype(np.float32)
preds_val_t = (preds_val > 0.5).astype(np.float32)
preds_test_t = (preds_test > 0.5).astype(np.float32)
train_acc = model.evaluate(X_train, Y_train, verbose=1)
test_acc = model.evaluate(X_test, Y_test, verbose=1)