def __init__(self, model_path, **config):
self.model_path = model_path
# Update config
self.config = dict_update(getattr(self, 'default_config', {}), config)
self._init_model()
if model_path is None:
print("No pretrained model specified!")
self.sess = None
else:
ext = os.path.splitext(model_path)[1]
sess_config = tf.compat.v1.ConfigProto()
sess_config.gpu_options.allow_growth = True
if ext.find('.pb') == 0:
graph = load_frozen_model(self.model_path, print_nodes=False)
self.sess = tf.compat.v1.Session(graph=graph, config=sess_config)
elif ext.find('.ckpt') == 0:
self.sess = tf.compat.v1.Session(config=sess_config)
meta_graph_path = os.path.join(model_path + '.meta')
if not os.path.exists(meta_graph_path):
self._construct_network()
recoverer(self.sess, model_path)
else:
recoverer(self.sess, model_path, meta_graph_path)
def main(argv=None): # pylint: disable=unused-argument
"""Program entrance."""
# create sift detector.
sift_wrapper = SiftWrapper(n_sample=FLAGS.max_kpt_num)
sift_wrapper.half_sigma = FLAGS.half_sigma
sift_wrapper.pyr_off = FLAGS.pyr_off
sift_wrapper.ori_off = FLAGS.ori_off
sift_wrapper.create()
# create deep feature extractor.
graph = load_frozen_model(FLAGS.model_path, print_nodes=False)
sess = tf.Session(graph=graph)
# extract deep feature from images.
deep_feat1, cv_kpts1, img1 = extract_deep_features(sift_wrapper,
sess,
FLAGS.img1_path,
qtz=False)
deep_feat2, cv_kpts2, img2 = extract_deep_features(sift_wrapper,
sess,
FLAGS.img2_path,
qtz=False)
# match features by OpenCV brute-force matcher (CPU).
matcher_wrapper = MatcherWrapper()
# the ratio criterion is set to 0.89 for GeoDesc as described in the paper.
deep_good_matches, deep_mask = matcher_wrapper.get_matches(
deep_feat1,
deep_feat2,
cv_kpts1,
cv_kpts2,
ratio=0.89,
cross_check=True,
info='deep')
deep_display = matcher_wrapper.draw_matches(img1, cv_kpts1, img2, cv_kpts2,
deep_good_matches, deep_mask)
# compare with SIFT.
if FLAGS.cf_sift:
sift_feat1 = sift_wrapper.compute(img1, cv_kpts1)
sift_feat2 = sift_wrapper.compute(img2, cv_kpts2)
sift_good_matches, sift_mask = matcher_wrapper.get_matches(
sift_feat1,
sift_feat2,
cv_kpts1,
cv_kpts2,
ratio=0.80,
cross_check=True,
info='sift')
sift_display = matcher_wrapper.draw_matches(img1, cv_kpts1, img2,
cv_kpts2,
sift_good_matches,
sift_mask)
display = np.concatenate((sift_display, deep_display), axis=0)
else:
display = deep_display
cv2.imshow('display', display)
cv2.waitKey()
sess.close()
def __init__(self, model_path, **config):
self.model_path = model_path
# Update config
self.config = dict_update(getattr(self, 'default_config', {}), config)
self._init_model()
ext = os.path.splitext(model_path)[1]
sess_config = tf.compat.v1.ConfigProto()
sess_config.gpu_options.allow_growth = True
if ext.find('.pb') == 0:
graph = load_frozen_model(self.model_path, print_nodes=False)
self.sess = tf.compat.v1.Session(graph=graph, config=sess_config)
elif ext.find('.ckpt') == 0:
self._construct_network()
self.sess = tf.compat.v1.Session(config=sess_config)
recoverer(self.sess, model_path)
def main(argv=None): # pylint: disable=unused-argument
"""Program entrance."""
hpatches_seq_list = open(FLAGS.hpatches_seq_list).read().splitlines()
n_seq = len(hpatches_seq_list)
graph = load_frozen_model(FLAGS.model_path, print_nodes=False)
with tf.Session(graph=graph) as sess:
for i in range(n_seq):
if i % 16 == 0:
print(i, '/', n_seq)
strs = hpatches_seq_list[i].split('/')
save_folder = os.path.join(FLAGS.feat_out_path, 'geodesc',
strs[-2])
if not os.path.exists(save_folder):
os.makedirs(save_folder)
seq_data = load_seq(
os.path.join(FLAGS.hpatches_root, hpatches_seq_list[i]))
feat = sess.run("squeeze_1:0", feed_dict={"input:0": seq_data})
csv_path = os.path.join(save_folder,
os.path.splitext(strs[-1])[0] + '.csv')
np.savetxt(csv_path, feat, delimiter=",", fmt='%.8f')
def main(argv=None): # pylint: disable=unused-argument
"""Program entrance."""
local_model_path = FLAGS.model_prefix + '.pb'
regional_model_path = FLAGS.regional_model
aug_model_path = FLAGS.model_prefix + '-aug.pb'
input_imgs = [FLAGS.img1_path, FLAGS.img2_path]
num_img = len(input_imgs)
# detect SIFT keypoints and crop image patches.
patches, npy_kpts, cv_kpts, sift_desc, imgs = get_kpt_and_patch(input_imgs)
# extract local features and keypoint matchability.
local_graph = load_frozen_model(local_model_path, print_nodes=False)
local_feat = []
l2_local_feat = []
kpt_m = []
with tf.Session(graph=local_graph) as sess:
for i in patches:
local_returns = sess.run(
["conv6_feat:0", "kpt_mb:0", "l2norm_feat:0"],
feed_dict={"input:0": np.expand_dims(i, -1)})
local_feat.append(local_returns[0])
kpt_m.append(local_returns[1])
l2_local_feat.append(local_returns[2])
tf.reset_default_graph()
# extract regional features.
regional_graph = load_frozen_model(regional_model_path, print_nodes=False)
regional_feat = []
with tf.Session(graph=regional_graph) as sess:
for i in imgs:
regional_returns = sess.run(
"res5c:0", feed_dict={"input:0": np.expand_dims(i, 0)})
regional_feat.append(regional_returns)
tf.reset_default_graph()
# local feature augmentation.
aug_graph = load_frozen_model(aug_model_path, print_nodes=False)
aug_feat = []
with tf.Session(graph=aug_graph) as sess:
for i in range(num_img):
aug_returns = sess.run("l2norm:0",
feed_dict={
"input/local_feat:0":
np.expand_dims(local_feat[i], 0),
"input/regional_feat:0":
regional_feat[i],
"input/kpt_m:0":
np.expand_dims(kpt_m[i], 0),
"input/kpt_xy:0":
np.expand_dims(npy_kpts[i], 0),
})
aug_returns = np.squeeze(aug_returns, axis=0)
aug_feat.append(aug_returns)
tf.reset_default_graph()
# feature matching and draw matches.
matcher = MatcherWrapper()
sift_match, sift_mask = matcher.get_matches(
sift_desc[0],
sift_desc[1],
cv_kpts[0],
cv_kpts[1],
ratio=0.8 if FLAGS.ratio_test else None,
cross_check=False,
err_thld=4,
info='sift')
base_match, base_mask = matcher.get_matches(
l2_local_feat[0],
l2_local_feat[1],
cv_kpts[0],
cv_kpts[1],
ratio=0.89 if FLAGS.ratio_test else None,
cross_check=False,
err_thld=4,
info='base')
aug_match, aug_mask = matcher.get_matches(
aug_feat[0],
aug_feat[1],
cv_kpts[0],
cv_kpts[1],
ratio=0.89 if FLAGS.ratio_test else None,
cross_check=False,
err_thld=4,
info='aug')
sift_disp = matcher.draw_matches(imgs[0], cv_kpts[0], imgs[1], cv_kpts[1],
sift_match, sift_mask)
base_disp = matcher.draw_matches(imgs[0], cv_kpts[0], imgs[1], cv_kpts[1],
base_match, base_mask)
aug_disp = matcher.draw_matches(imgs[0], cv_kpts[0], imgs[1], cv_kpts[1],
aug_match, aug_mask)
rows, cols = sift_disp.shape[0:2]
white = (np.ones((rows / 50, cols, 3)) * 255).astype(np.uint8)
disp = np.concatenate([sift_disp, white, base_disp, white, aug_disp],
axis=0)
plt.xticks([])
plt.yticks([])
plt.imshow(disp)
plt.show()