-
Notifications
You must be signed in to change notification settings - Fork 1
/
simple.py
359 lines (279 loc) · 10.9 KB
/
simple.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
355
356
357
358
359
# -*- coding: utf-8 -*-
"""
Created on Fri Mar 1 11:03:45 2019
@author: jwang
"""
# -*- coding: utf-8 -*-
"""
Created on Thu Feb 28 13:36:09 2019
@author: jwang
"""
## test 2, version 2, trained by all three "hsv" channels of the images, and the most simplifed CNN
import json
import pandas as pd
import matplotlib.pyplot as plt
import cv2
#%%
#load training data
"""
# as mentioned, there are 5 images in which there are no corners (totally blocked by pillar) and the ground truth
in JSON are [[ ]]. These images are IMG_0492.JPG, IMG_0688.JPG, IMG_1343.JPG, IMG_5199.JPG, and IMG_6608.JPG.
There are another 11 images was half blocked and thus the corresponding ground truth labels can be considered
as errors including IMG_0493.JPG, IMG_1342.JPG, IMG_1344.JPG, IMG_5196.JPG, IMG_5197.JPG, IMG_5198.JPG, IMG_5200.JPG,
IMG_5201.JPG, IMG_5202.JPG, IMG_6607.JPG, and IMG_6609.JPG. For those above images in the './AlphaPilot_test2/Data_Training/'
and corresponding ground truth in './AlphaPilot_test2/training_GT_labels_v2.json' were removed before loading.
"""
json_data = json.load(open("C:/Users/jwang/Desktop/AlphaPilot_test2/augmented_training_GT_labels_v2.json"))
labels_number = pd.read_json(json.dumps(json_data), typ='frame', orient="index")
labels_img = pd.read_json(json.dumps(json_data), typ='series')
#print(len(labels_number))
#print(labels_number[0][1])
#print(labels_img[0][0])
data_folder = 'C:/Users/jwang/Desktop/AlphaPilot_test2/Data_Training/'
#%%
# preprocess the image data by resizing
# preprocess the image data by changing the image color space from RGB to HSV and keep S channel only for light rubustness
def image_preprocessing(img):
resized = cv2.resize((cv2.cvtColor(img, cv2.COLOR_RGB2HSV)),(img_cols,img_rows))
#gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
#resized = cv2.resize(gray, (img_cols,img_rows))
return resized
# normalize the training images
def normalized(img):
#return (img - 128) / 128
# normalize image data to [0.1, 0.9]
a = 0
b = 1.0
minimum = 0
maximum = 255
return (b - a) * ((img - minimum) / (maximum - minimum)) + a
# resize the images from 1296*864 to 96*96
# the image dimension is tunable
img_rows = 96
img_cols = 96
#%%
import numpy as np
# X for images as training features and y for coordinates as training labels
X = []
y = []
N_test = 256*2
#for i in range(N_test):
for i in range(len(labels_number)):
img_path = labels_number.index[i]
img_path = data_folder+img_path
print("loading images:" +img_path)
img = plt.imread(img_path)
X.append(image_preprocessing(img))
y.append(labels_number[0][i])
print(len(y))
#plt.imshow(img)
#plt.scatter(y[-1][0:2], y[-1][3:5], s=img)
#plt.show()
#%%
# plot testing
"""
img_test = plt.imread('C:/Users/jwang/Desktop/AlphaPilot_test2/error_data/IMG_0493.JPG')
plt.scatter(x=[729, 794, 789, 724], y=[330, 331, 492, 492], c='r', s=40)
#plt.plot([729, 330, 794, 331, 789, 492, 724, 492],'o')
plt.imshow(img_test)
#%%
img_test = plt.imread('C:/Users/jwang/Desktop/AlphaPilot_test2/Data_Training/IMG_0640.JPG')
coord = [563, 228, 729, 115, 732, 743, 558, 634]
plt.scatter(x=[coord[0], coord[2], coord[4], coord[6]], y=[coord[1], coord[3], coord[5], coord[7]], c='r', s=40)
#plt.plot([729, 330, 794, 331, 789, 492, 724, 492],'o')
plt.imshow(img_test)
"""
#%%
from sklearn.utils import shuffle
data={}
data['features'] = X
data['labels'] = y
X_train, y_train = data['features'], data['labels']
X_train = np.asarray(data['features']).astype('float32')
y_train = np.asarray(data['labels']).astype('float32')
# normalize the training label
y_train = (y_train - 48) / 48 # scale target coordinates to [-1, 1]
# shuffle data and then split data into training and validation
X_train, y_train = shuffle(X_train, y_train, random_state=42)
# for tensorflow
# from sklearn.model_selection import train_test_split
#X_train, X_valid, y_train, y_valid = train_test_split(X_train, y_train, random_state=0, test_size=0.1)
#%%
# reshape the training data to (None, img_rows, img_cols, 1)
X_train = X_train.reshape(X_train.shape[0], img_rows, img_cols, 3)
#X_valid = X_valid.reshape(X_valid.shape[0], img_rows, img_cols, 1)
#y_train = y_train.reshape(y_train.shape[0], 8)
#y_valid = y_valid.reshape(y_valid.shape[0], 8)
#%%
X_train = normalized(X_train)
#X_valid = normalized(X_valid)
print("Training Image Shape: {}".format(X_train[0].shape))
print("Training Set: {} samples".format(len(X_train)))
print()
print("Training Label Shape: {}".format(y_train[0].shape))
print("Training Set: {} samples".format(len(y_train)))
#print("Validation Image Shape: {}".format(X_valid[0].shape))
#print("Validation Set: {} samples".format(len(X_valid)))
#%%
# https://github.com/elix-tech/kaggle-facial-keypoints/blob/master/kfkd.py
# Import Keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Convolution2D, MaxPooling2D, Dropout, Activation
from tensorflow.keras.layers import Flatten, Dense
model = Sequential()
model.add(Convolution2D(32, 3, 3, input_shape=(96, 96, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.1))
model.add(Convolution2D(64, 2, 2))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Convolution2D(128, 2, 2))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(1000))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(1000))
model.add(Activation('relu'))
model.add(Dense(8))
model.summary()
#%%
"""
# this is a shallow CNN built bu keras I directly grabed from my previous project. A deeper CNN architecture
can definetely be tested, but I feel since some of the ground truth values are not very accurate made on purpose,
a shallow CNN maybe works better.
# the model accept 96x96 pixel one chanel images in
# a fully-connected output layer with 8 values (2 for each corner)
model = Sequential()
model.add(Convolution2D(32, (5, 5), input_shape=(96,96,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Convolution2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Convolution2D(128, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Convolution2D(15, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(256, activation='relu'))
model.add(Dense(128, activation='relu'))
model.add(Dense(64, activation='relu'))
model.add(Dense(8))
# Summarize the model
model.summary()
"""
#%%
from tensorflow.keras.optimizers import SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, Nadam
# Compile the model
model.compile(optimizer='adam', loss='mean_squared_error', metrics=['accuracy'])
# Train the model
hist = model.fit(X_train, y_train, epochs=200, batch_size=128, verbose=1, validation_split=0.15)
# Save the model as model.h5
model.save('my_model_3c_simple2_arg.h5')
print("Trained model saved!")
#%%
# training history
print(hist.history.keys())
# summarize history for accuracy
plt.plot(hist.history['acc'])
plt.plot(hist.history['val_acc'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
# summarize history for loss
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
#%%
# RESIZE images
import cv2
filename = 'C:/Users/JWang/Dropbox/AAA_PostDoc/AlphaPilot/RESULTES/simple_arg_2timesTraining/test10.jpg'
oriimg = plt.imread(filename)
newimg = cv2.resize(oriimg,(1296, 864))
cv2.imshow("Show by CV2",newimg)
cv2.waitKey(0)
image = cv2.imwrite('C:/Users/JWang/Dropbox/AAA_PostDoc/AlphaPilot/RESULTES/simple_arg_2timesTraining/test10.jpg',255*newimg)
#%%
# testing the trained CNN by images in './Data_LeaderboardTesting'
from tensorflow.keras.models import load_model
img_rows = 96
img_cols = 96
model = load_model('/home/aml/Dropbox/AAA_PostDoc/AlphaPilot/RESULTES/simple_arg_2timesTraining/my_model_3c.h5')
X_test = plt.imread('/home/aml/Dropbox/AAA_PostDoc/AlphaPilot/RESULTES/simple_arg_2timesTraining/test8.jpg')
X_test = cv2.resize((cv2.cvtColor(X_test, cv2.COLOR_RGB2HSV)),(img_cols,img_rows))
#X_test = cv2.resize(cv2.cvtColor(X_test, cv2.COLOR_RGB2GRAY),(img_cols,img_rows))
X_test = X_test / 255
X_test = X_test.reshape(-1, img_rows, img_cols, 3)
y_test = model.predict(X_test)
X_org = plt.imread('/home/aml/Dropbox/AAA_PostDoc/AlphaPilot/RESULTES/simple_arg_2timesTraining/test8.jpg')
y_test = y_test * 48 + 48 # undo the normalization
plt.scatter(x=[y_test[0][0], y_test[0][2], y_test[0][4], y_test[0][6]], y=[y_test[0][1], y_test[0][3], y_test[0][5], y_test[0][7]], c='r', s=40)
plt.imshow(X_org)
#%%
# Load libraries
import json
from pprint import pprint
import glob
import cv2
import numpy as np
from random import shuffle
from helper import *
import time
import os
import json
import cv2
import numpy as np
from tensorflow.keras import models
import pickle
#%%
model = load_model('my_model_3c_simple2_arg.h5')
def img_predict(img):
n_boxes = 1
if n_boxes>0:
bb_all = 400*np.random.uniform(size = (n_boxes,9))
bb_all[:, 0:8] = model.predict(img)
bb_all = bb_all * 48 + 48
bb_all[:,-1] = 0.5
else:
bb_all = []
return bb_all.tolist()
img_file = glob.glob('Data_LeaderboardTesting/*.JPG')
img_keys = [img_i.split('/')[-1] for img_i in img_file]
# load image, convert to RGB, run model and plot detections.
time_all = []
pred_dict = {}
for img_key in img_keys:
img =cv2.imread(img_key)
img_org =cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img = cv2.resize((cv2.cvtColor(img_org, cv2.COLOR_BGR2RGB)),(96, 96))
img = img / 255
img = img.reshape(-1, 96, 96, 3)
tic = time.monotonic()
bb_all = img_predict(img)
toc = time.monotonic()
pred_dict[os.path.basename(img_key)] = bb_all # http://techs.studyhorror.com/python-get-last-directory-name-in-path-i-139
time_all.append(toc-tic)
plot_bbox(img_org, bb_all)
#%%
mean_time = np.mean(time_all)
ci_time = 1.96*np.std(time_all)
freq = np.round(1/mean_time,2)
print('95% confidence interval for inference time is {0:.2f} +/- {1:.4f}.'.format(mean_time,ci_time))
print('Operating frequency from loading image to getting results is {0:.2f}.'.format(freq))
#%%
import json
with open('random_submission.json', 'w') as f:
json.dump(pred_dict, f)