CNN FINETUNING WITH PRE-TRAINED VGG NET
import os
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
import scipy.misc
import scipy.io
from tensorflow.examples.tutorials.mnist import input_data
%matplotlib inline
print ("Packages loaded.")
Packages loaded.
LOAD DATA
cwd = os.getcwd()
loadpath = cwd + "/data/data4vgg.npz"
l = np.load(loadpath)
print (l.files)
['trainlabel', 'trainimg', 'testimg', 'testlabel']
PARSE DATA
trainimg = l['trainimg']
trainlabel = l['trainlabel']
testimg = l['testimg']
testlabel = l['testlabel']
ntrain = trainimg.shape[0]
nclass = trainlabel.shape[1]
dim = trainimg.shape[1]
ntest = testimg.shape[0]
print ("%d train images loaded" % (ntrain))
print ("%d test images loaded" % (ntest))
print ("%d dimensional input" % (dim))
print ("%d classes" % (nclass))
print ("shape of 'trainimg' is %s" % (trainimg.shape,))
print ("shape of 'testimg' is %s" % (testimg.shape,))
69 train images loaded
18 test images loaded
37632 dimensional input
2 classes
shape of 'trainimg' is (69, 37632)
shape of 'testimg' is (18, 37632)
GENERATE TENSORS FOR TRAINING AND TESTING
trainimg_tensor = np.ndarray((ntrain, 112, 112, 3))
for i in range(ntrain):
currimg = trainimg[i, :]
currimg = np.reshape(currimg, [112, 112, 3])
trainimg_tensor[i, :, :, :] = currimg
print ("shape of trainimg_tensor is %s" % (trainimg_tensor.shape,))
testimg_tensor = np.ndarray((ntest, 112, 112, 3))
for i in range(ntest):
currimg = testimg[i, :]
currimg = np.reshape(currimg, [112, 112, 3])
testimg_tensor[i, :, :, :] = currimg
print ("shape of testimg_tensor is %s" % (testimg_tensor.shape,))
shape of trainimg_tensor is (69, 112, 112, 3)
shape of testimg_tensor is (18, 112, 112, 3)
DEFINE A FUNCTION FOR USING PRETRAINED VGG NETWORK
def net(data_path, input_image):
layers = (
'conv1_1', 'relu1_1', 'conv1_2', 'relu1_2', 'pool1',
'conv2_1', 'relu2_1', 'conv2_2', 'relu2_2', 'pool2',
'conv3_1', 'relu3_1', 'conv3_2', 'relu3_2', 'conv3_3',
'relu3_3', 'conv3_4', 'relu3_4', 'pool3',
'conv4_1', 'relu4_1', 'conv4_2', 'relu4_2', 'conv4_3',
'relu4_3', 'conv4_4', 'relu4_4', 'pool4',
'conv5_1', 'relu5_1', 'conv5_2', 'relu5_2', 'conv5_3',
'relu5_3', 'conv5_4', 'relu5_4'
)
data = scipy.io.loadmat(data_path)
mean = data['normalization'][0][0][0]
mean_pixel = np.mean(mean, axis=(0, 1))
weights = data['layers'][0]
net = {}
current = input_image
for i, name in enumerate(layers):
kind = name[:4]
if kind == 'conv':
kernels, bias = weights[i][0][0][0][0]
kernels = np.transpose(kernels, (1, 0, 2, 3))
bias = bias.reshape(-1)
current = _conv_layer(current, kernels, bias)
elif kind == 'relu':
current = tf.nn.relu(current)
elif kind == 'pool':
current = _pool_layer(current)
net[name] = current
assert len(net) == len(layers)
return net, mean_pixel
def _conv_layer(input, weights, bias):
conv = tf.nn.conv2d(input, tf.constant(weights), strides=(1, 1, 1, 1),
padding='SAME')
return tf.nn.bias_add(conv, bias)
def _pool_layer(input):
return tf.nn.max_pool(input, ksize=(1, 2, 2, 1), strides=(1, 2, 2, 1),
padding='SAME')
def preprocess(image, mean_pixel):
return image - mean_pixel
def unprocess(image, mean_pixel):
return image + mean_pixel
VGG_PATH = cwd + "/data/imagenet-vgg-verydeep-19.mat"
with tf.Graph().as_default(), tf.Session() as sess:
with tf.device("/cpu:0"):
img_placeholder = tf.placeholder(tf.float32, shape=(None, 112, 112, 3))
net_val, mean_pixel = net(VGG_PATH, img_placeholder)
train_features = net_val['relu5_4'].eval(feed_dict={img_placeholder: trainimg_tensor})
test_features = net_val['relu5_4'].eval(feed_dict={img_placeholder: testimg_tensor})
print ("TYPE OF 'train_features' IS %s" % (type(train_features)))
print ("SHAPE OF 'train_features' IS %s" % (train_features.shape,))
print ("TYPE OF 'test_features' IS %s" % (type(test_features)))
print ("SHAPE OF 'test_features' IS %s" % (test_features.shape,))
print("PREPROCESSING DONE")
TYPE OF 'train_features' IS <type 'numpy.ndarray'>
SHAPE OF 'train_features' IS (69, 7, 7, 512)
TYPE OF 'test_features' IS <type 'numpy.ndarray'>
SHAPE OF 'test_features' IS (18, 7, 7, 512)
PREPROCESSING DONE
VECTORIZE CNN FEATURES
train_vectorized = np.ndarray((ntrain, 7*7*512))
test_vectorized = np.ndarray((ntest, 7*7*512))
for i in range(ntrain):
curr_feat = train_features[i, :, :, :]
curr_feat_vec = np.reshape(curr_feat, (1, -1))
train_vectorized[i, :] = curr_feat_vec
for i in range(ntest):
curr_feat = test_features[i, :, :, :]
curr_feat_vec = np.reshape(curr_feat, (1, -1))
test_vectorized[i, :] = curr_feat_vec
print ("SHAPE OF 'train_vectorized' IS %s" % (train_vectorized.shape,))
print ("SHAPE OF 'test_vectorized' IS %s" % (test_vectorized.shape,))
SHAPE OF 'train_vectorized' IS (69, 25088)
SHAPE OF 'test_vectorized' IS (18, 25088)
DEFINE NETWORKS AND FUNCTIONS (ADD 2LAYER MLP)
learning_rate = 0.0001
training_epochs = 100
batch_size = 100
display_step = 10
with tf.device("/cpu:0"):
n_input = dim
n_output = nclass
weights = {
'wd1': tf.Variable(tf.random_normal([7*7*512, 1024], stddev=0.1)),
'wd2': tf.Variable(tf.random_normal([1024, n_output], stddev=0.1))
}
biases = {
'bd1': tf.Variable(tf.random_normal([1024], stddev=0.1)),
'bd2': tf.Variable(tf.random_normal([n_output], stddev=0.1))
}
def conv_basic(_input, _w, _b, _keepratio):
_input_r = _input
_dense1 = tf.reshape(_input_r, [-1, _w['wd1'].get_shape().as_list()[0]])
_fc1 = tf.nn.relu(tf.add(tf.matmul(_dense1, _w['wd1']), _b['bd1']))
_fc_dr1 = tf.nn.dropout(_fc1, _keepratio)
_out = tf.add(tf.matmul(_fc_dr1, _w['wd2']), _b['bd2'])
out = {'input_r': _input_r, 'dense1': _dense1,
'fc1': _fc1, 'fc_dr1': _fc_dr1, 'out': _out }
return out
x = tf.placeholder(tf.float32, [None, 7*7*512])
y = tf.placeholder(tf.float32, [None, n_output])
keepratio = tf.placeholder(tf.float32)
with tf.device("/cpu:0"):
_pred = conv_basic(x, weights, biases, keepratio)['out']
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(_pred, y))
optm = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
_corr = tf.equal(tf.argmax(_pred,1), tf.argmax(y,1))
accr = tf.reduce_mean(tf.cast(_corr, tf.float32))
init = tf.initialize_all_variables()
print ("Network Ready to Go!")
Network Ready to Go!
CNN FINETUNING
sess = tf.Session()
sess.run(init)
for epoch in range(training_epochs):
avg_cost = 0.
num_batch = int(ntrain/batch_size)+1
for i in range(num_batch):
randidx = np.random.randint(ntrain, size=batch_size)
batch_xs = train_vectorized[randidx, :]
batch_ys = trainlabel[randidx, :]
sess.run(optm, feed_dict={x: batch_xs, y: batch_ys, keepratio:0.7})
avg_cost += sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keepratio:1.})/num_batch
if epoch % display_step == 0:
print ("Epoch: %03d/%03d cost: %.9f" % (epoch, training_epochs, avg_cost))
train_acc = sess.run(accr, feed_dict={x: batch_xs, y: batch_ys, keepratio:1.})
print (" Training accuracy: %.3f" % (train_acc))
test_acc = sess.run(accr, feed_dict={x: test_vectorized, y: testlabel, keepratio:1.})
print (" Test accuracy: %.3f" % (test_acc))
print ("Optimization Finished!")
Epoch: 000/100 cost: 3.248429298
Training accuracy: 0.760
Test accuracy: 0.444
Epoch: 010/100 cost: 0.146418720
Training accuracy: 0.980
Test accuracy: 0.722
Epoch: 020/100 cost: 0.000040202
Training accuracy: 1.000
Test accuracy: 0.778
Epoch: 030/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.778
Epoch: 040/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.778
Epoch: 050/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.778
Epoch: 060/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.722
Epoch: 070/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.722
Epoch: 080/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.778
Epoch: 090/100 cost: 0.000000000
Training accuracy: 1.000
Test accuracy: 0.889
Optimization Finished!