def main(path):
cars, noncars, data_info = load_training_data(path)
ind = np.random.randint(0, len(cars))
name = cars[ind]
img = cv2.imread(name)
# Select a small fraction of pixels to plot by subsampling it
scale = max(img.shape[0], img.shape[1], 64) / 64 # at most 64 rows and columns
img_small = cv2.resize(img, (np.int(
img.shape[1] / scale), np.int(img.shape[0] / scale)), interpolation=cv2.INTER_NEAREST)
# Convert subsampled image to desired color space(s)
# OpenCV uses BGR, matplotlib likes RGB
img_small_RGB = cv2.cvtColor(img_small, cv2.COLOR_BGR2RGB)
img_small_HSV = cv2.cvtColor(img_small, cv2.COLOR_BGR2HSV)
img_small_rgb = img_small_RGB / 255. # scaled to [0, 1], only for plotting
# Plot and show
plot3d(img_small_RGB, img_small_rgb)
plt.suptitle(name)
plt.show()
plot3d(img_small_HSV, img_small_rgb, axis_labels=list("HSV"))
plt.suptitle(name)
plt.show()
def get_clean_data():
x, y = load_training_data()
# Reshape images from 1x4096 to 64x64
x = threshold(x)
x = remove_dots(x)
x = x.reshape(-1, 64, 64, 1)
test_split = 0.1
np.random.seed(113)
indices = np.arange(len(x))
np.random.shuffle(indices)
_, num_to_index, _ = get_labels()
x = x[indices]
y = y[indices]
y = [num_to_index[yi] for yi in y.tolist()]
x_train = np.array(x[:int(len(x) * (1 - test_split))])
y_train = np.array(y[:int(len(x) * (1 - test_split))])
x_test = np.array(x[int(len(x) * (1 - test_split)):])
y_test = np.array(y[int(len(x) * (1 - test_split)):])
return (x_train, y_train), (x_test, y_test)
def train(num_iters):
train_data, train_labels = load_training_data()
with tf.Graph().as_default():
with tf.Session() as sess:
batch_inputs,batch_labels=create_placeholders(100,32,32,3,10)
logits=vgg.inference_vgg(batch_inputs)
loss=vgg.loss(logits,batch_labels)
train_op=vgg.training(loss,0.0001)
saver = tf.train.Saver(tf.all_variables())
#load most recent checkpoint
ckpt=tf.train.get_checkpoint_state('./checkpoints')
if ckpt:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
global_step=int(global_step)
else:
print('No checkpoint found')
global_step=0
#summary_op=tf.merge_all_summaries() #causes problems with placeholders
#summary_writer=tf.train.SummaryWriter('log',sess.graph)
#if (global_step==0):
if (global_step==0):
sess.run(tf.initialize_all_variables())
for iter in range(num_iters):
global_step+=1
start_time=time.time()
inps,labs=create_batch(iters,train_data,train_labels)
feed_dict={
batch_inputs: inps,
batch_labels: labs
}
_, total_loss = sess.run([train_op,loss],feed_dict=feed_dict)
end_time=time.time()
total_time=(end_time-start_time)
if (global_step%5)==0:
print('Iteration: %d, Loss: %.2f, Iteration time: %.1f' %(global_step,total_loss,total_time))
#if (iter%50==0):
# summary_str=sess.run(summary_op)
# summary_writer.add_summary(summary_str,iter)
if (global_step%1000==0):
checkpoint_path = os.path.join('./checkpoints', 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=global_step)
def get_training_data():
print('\ngetting training data')
x_train, y_train = load_training_data()
Y_train = np_utils.to_categorical(y_train, len(LABELS))
X_val, Y_val = get_validation_data()
return x_train, Y_train, X_val, Y_val
