learning_rate = 0.001
beta = 4
n_disp = 5
n_disp_validation = 20
loss_type = 'min'
output_type = 'axis'
log_dir = os.path.join(args.logdir, args.name)
if not tf.gfile.Exists(log_dir):
tf.gfile.MakeDirs(log_dir)
if not tf.gfile.Exists(args.save_dir):
tf.gfile.MakeDirs(args.save_dir)
# Prepare input queues
train_reader = ImageReader(args.dataset, batch_size=args.batch_size,
image_size=image_size,
randomise=True, augment=True, centre_crop=False,
convert_to_axis=(output_type=='axis'))
if args.validate:
validation_reader = ImageReader(args.validate, batch_size=args.batch_size,
image_size=image_size, randomise=True,
centre_crop=False, convert_to_axis=(output_type=='axis'))
# tf Graph input
k = 6 if output_type == 'axis' else 7
x = tf.placeholder(tf.float32, [None, None, None, 3], name="InputData")
y = tf.placeholder(tf.float32, [None, k], name="LabelData")
# Define the network
poseNet = Posenet(endpoint='Mixed_5b', n_fc=256, loss_type=loss_type, output_type=output_type)
train_output, train_loss, train_summaries = poseNet.create_trainable(x, y, beta=beta, learn_beta=True, dropout=0.5)
if args.validate:
import argparse
from time import sleep
import matplotlib.pyplot as plt
from posenet.core.image_reader import ImageReader
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', action='store', required=True)
args = parser.parse_args()
input_size = 224
crop_size = 200
reader = ImageReader(args.dataset, batch_size=16,
image_size=[input_size, input_size],
randomise=True, augment=True, crop_size=[crop_size, crop_size])
fig = plt.figure(num=None, figsize=(6, 6), dpi=80, facecolor='w')
image, label = reader.next_batch()
for j in range(16):
fig.add_subplot(4, 4, j+1)
plt.imshow((1+image[j])/2.0, aspect='auto')
plt.axis('off')
plt.subplots_adjust(hspace=0, wspace=0)
plt.show()
help='''Connect consecutive camera positions with lines''')
parser.add_argument('--rings', action='store', nargs='*', required=False)
parser.add_argument('--plot_gt', action='store_true')
parser.add_argument('--plot_diff', action='store_true')
parser.add_argument('-u', '--uncertainty', action='store_true')
args = parser.parse_args()
if not args.model or args.plot_gt or args.plot_diff:
_, labels = read_label_file(args.dataset)
positions_gt = np.array([l[0:3] for l in labels])
orientations_gt = np.array([l[3:7] for l in labels])
if args.model:
input_size = 256
test_reader = ImageReader(args.dataset,
batch_size=1,
image_size=[input_size, input_size],
randomise=False)
n_images = test_reader.total_images()
positions = np.empty([0, 3])
orientations = np.empty([0, 4])
if args.uncertainty:
std_x = []
std_q = []
with Localiser(args.model, uncertainty=args.uncertainty) as localiser:
for i in range(n_images):
images_feed, labels_feed = test_reader.next_batch()
# Make prediction
predicted = localiser.localise(images_feed)
n_input = 224
learning_rate = 0.001
beta = 4
n_disp = 5
n_disp_validation = 20
log_dir = os.path.join(args.logdir, args.name)
if not tf.gfile.Exists(log_dir):
tf.gfile.MakeDirs(log_dir)
if not tf.gfile.Exists(args.save_dir):
tf.gfile.MakeDirs(args.save_dir)
# Prepare input queues
train_reader = ImageReader(args.dataset, batch_size=args.batch_size,
image_size=[n_input, n_input],
random_crop=False, randomise=True,
augment=True)
if args.validate:
validation_reader = ImageReader(args.validate, batch_size=args.batch_size,
image_size=[n_input, n_input],
random_crop=False, randomise=True)
# tf Graph input
x = tf.placeholder(tf.float32, [None, n_input, n_input, 3], name="InputData")
y = tf.placeholder(tf.float32, [None, 7], name="LabelData")
# Define the network
poseNet = Posenet(endpoint='Mixed_5b', n_fc=256)
train_output, train_loss, train_summaries = poseNet.create_trainable(x, y, beta=beta, learn_beta=True)
if args.validate:
validation_output, validation_loss, validation_summaries = poseNet.create_validation(x, y)
from posenet.utils import rotate_by_quaternion
output_type = 'axis' if args.axis else 'quat'
if not args.model or args.plot_gt or args.plot_diff:
_, labels = read_label_file(args.dataset,
convert_to_axis=args.convert_to_axis)
positions_gt = np.array([l[0:3] for l in labels])
orientations_gt = np.array([l[3:] for l in labels])
if args.model:
input_size = [256, 455]
crop_size = [256, 256]
test_reader = ImageReader(args.dataset,
batch_size=1,
image_size=input_size,
crop_size=crop_size,
centre_crop=True,
randomise=False)
n_images = test_reader.total_images()
positions = []
orientations = []
if args.uncertainty:
std_x = []
std_q = []
with Localiser(args.model,
uncertainty=args.uncertainty,
output_type=output_type,
dropout=0.3) as localiser:
for i in range(n_images):
import argparse
from time import sleep
import matplotlib.pyplot as plt
from posenet.core.image_reader import ImageReader
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', action='store', required=True)
args = parser.parse_args()
input_size = 224
reader = ImageReader(args.dataset,
batch_size=16,
image_size=[input_size, input_size],
random_crop=False,
randomise=True,
augment=True,
normalise=False)
fig = plt.figure(num=None, figsize=(6, 6), dpi=80, facecolor='w')
image, label = reader.next_batch()
for j in range(16):
fig.add_subplot(4, 4, j + 1)
plt.imshow(image[j], aspect='auto')
plt.axis('off')
plt.subplots_adjust(hspace=0, wspace=0)
plt.show()