/
train_accel.py
125 lines (97 loc) · 5.17 KB
/
train_accel.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
import os
import tensorflow as tf
from tensorflow.core.protobuf import saver_pb2
import driving_data
import scipy
import cv2
# standard step - reset computation graphs
tf.reset_default_graph()
saveDirectory = './save'
L2NormConst = 0.001
import model_accel
all_vars = tf.trainable_variables()
# model_accel_vars = [k for k in all_vars if 'accel' in k.name]
print(model_accel.y_accel_.shape)
print(model_accel.y_accel.shape)
loss_accel = tf.reduce_mean(tf.square(tf.subtract(model_accel.y_accel_, model_accel.y_accel))) #+ tf.add_n([tf.nn.l2_loss(v) for v in all_vars]) * L2NormConst
train_step_accel = tf.train.AdadeltaOptimizer(1., 0.95, 1e-6).minimize(loss_accel) # These are default parameters to the Adadelta optimizer
# train_step_accel = tf.train.AdamOptimizer().minimize(loss_accel)
sess_accel = tf.Session()
sess_accel.run(tf.global_variables_initializer())
saver_accel = tf.train.Saver(all_vars)
# create a summary to monitor cost tensor
tf.summary.scalar("loss_accel", loss_accel)
# merge all summaries into a single op
merged_summary_op = tf.summary.merge_all()
# op to write logs to Tensorboard
logs_path = './logs'
summary_writer = tf.summary.FileWriter(logs_path, graph=tf.get_default_graph())
epochs = 50
batch_size = 80
# train over the dataset about 30 times
for epoch in range(epochs):
for i in range(int(driving_data.num_images/batch_size)):
xs, ys, accels, brakes = driving_data.LoadTrainBatch(batch_size, False)
train_step_accel.run(feed_dict={model_accel.x_accel: xs, model_accel.y_accel_: accels, model_accel.keep_prob_accel: 0.5, model_accel.keep_prob_accel_conv: 0.25}, session = sess_accel)
if i % 10 == 0:
xs, ys, accels, brakes = driving_data.LoadValBatch(batch_size, False)
loss_value_accel = loss_accel.eval(feed_dict={model_accel.x_accel:xs, model_accel.y_accel_: accels, model_accel.keep_prob_accel: 1.0, model_accel.keep_prob_accel_conv: 1.0}, session = sess_accel)
print("Epoch: %d, Step: %d, Accel Loss: %g " % (epoch, epoch * batch_size + i, loss_value_accel))
if i % batch_size == 0:
if not os.path.exists(saveDirectory):
os.makedirs(saveDirectory)
accel_checkpoint_path = os.path.join(saveDirectory, "model_accel.ckpt")
filename_accel = saver_accel.save(sess_accel, accel_checkpoint_path)
print("Model saved in file: %s" % filename_accel)
### To predict the output based on the above trained model###
xs = []
dataPath = "indian_dataset/"
fileNamePrefix = "circuit2_x264.mp4 "
with open(dataPath+"data.txt") as f:
for line in f:
xs.append(dataPath + fileNamePrefix + str(int(line.split()[0])).zfill(5)+".jpg")
i = 0
while(cv2.waitKey(10) != ord('q')):
full_image = scipy.misc.imread(xs[i], mode="RGB")
image = scipy.misc.imresize(full_image[-150:], [112, 112]) / 255.0
# print(image.shape)
# print(model.y.eval(feed_dict={model.x: [image], model.keep_prob: 1.0}))
# with g_accel.as_default():
acceleration = model_accel.y_accel.eval(feed_dict={model_accel.x_accel: [image], model_accel.keep_prob_accel: 1.0, model_accel.keep_prob_accel_conv: 1.0}, session = sess_accel)[0][0]
print(i,acceleration * 180.0 / scipy.pi)
i += 1
# To Visualize CNN Layers to better interpretability. Base code obtained from:
# https://medium.com/@awjuliani/visualizing-neural-network-layer-activation-tensorflow-tutorial-d45f8bf7bbc4
# def getActivations(layer,stimuli, filename, steer, acceleration):
# units = sess.run(layer,feed_dict={model.x:np.reshape(stimuli,[-1, 136, 240, 3],order='F'), model.y1_: np.reshape(steer, [-1, 1],order='F'), model.y2_: acceleration, model.keep_prob:1.0})
# plotNNFilter(units, filename)
# def plotNNFilter(units, filename):
# filters = units.shape[3]
# plt.figure(1, figsize=(20,20))
# n_columns = 6
# n_rows = math.ceil(filters / n_columns) + 1
# for i in range(filters):
# plt.subplot(n_rows, n_columns, i+1)
# plt.title('Filter ' + str(i))
# plt.imshow(units[0,:,:,i], interpolation="nearest", cmap="gray")
# plt.savefig(filename)
# xs = []
# ys = []
# accels = []
# with open("indian_dataset/data.txt") as f:
# for line in f:
# xs.append("indian_dataset/circuit2_x264.mp4 " + str(line.split()[0]).zfill(5) + ".jpg")
# ys.append(float(line.split()[1]) * scipy.pi / 180)
# accels.append(float(line.split()[2]))
# frameNum = 4805
# full_image = scipy.misc.imread(xs[frameNum], mode="RGB")
# image = scipy.misc.imresize(full_image[-150:], [136, 240]) / 255.0
# cv2.imshow("Visualize CNN: input image", cv2.cvtColor(full_image, cv2.COLOR_RGB2BGR))
# cv2.imshow("Visualize CNN: input image", image)
# getActivations(model_steer.h_conv1,image, 'cnn-depict-conv1.jpg', ys[frameNum], accels[frameNum])
# getActivations(model_steer.h_conv2,image, 'cnn-depict-conv2.jpg', ys[frameNum], accels[frameNum])
# getActivations(model_steer.h_conv3,image, 'cnn-depict-conv3.jpg', ys[frameNum], accels[frameNum])
# getActivations(model_steer.h_conv4,image, 'cnn-depict-conv4.jpg', ys[frameNum], accels[frameNum])
# getActivations(model_steer.h_conv5,image, 'cnn-depict-conv5.jpg', ys[frameNum], accels[frameNum])
# getActivations(model_steer.h_conv6,image, 'cnn-depict-conv6.jpg', ys[frameNum], accels[frameNum])
sess_accel.close()