def extract_descriptor(self, image, feature):
img = np.array(fu.all_to_gray(image), dtype='float32')
_, desc, _ = self.fe.run(img)
if desc is not None:
desc = desc.T
return desc
def extract_descriptor(self, image, feature):
image = feature_utils.all_to_gray(image)
image = image / 255.0
feature, descriptor = eng.vl_sift(
matlab.single(image.tolist()), 'Magnif', self.Magnif,
nargout=2) # , peak_thresh=self.peak_thresh
descriptor = np.transpose(np.array(descriptor))
return descriptor
def extract_all(self, image):
img = np.array(fu.all_to_gray(image), dtype='float32')
pts, desc, heatmap = self.fe.run(img)
pts = pts[:2, :].T
if desc is not None:
desc = desc.T
return (pts, desc)
def extract_descriptor(self, image, feature):
gray_image = utils.all_to_gray(image)
patches = []
for f in feature:
patch = utils.extract_patch(gray_image, f, patch_sz=64)
patches.append(patch)
patches = np.array(patches)
desc = self.extract_descriptors_from_patch_batch(patches)
return desc
def detect_feature(self, image):
img = np.array(fu.all_to_gray(image), dtype='float32') / 255.0
pts, _, _ = self.fe.run(img)
nb_kpts = MAX_CV_KPTS
#Already ordered by confidence
if pts.shape[1] < MAX_CV_KPTS:
nb_kpts = pts.shape[1]
pts = pts[:2, :nb_kpts].T
return pts
def extract_all(self, image):
# pdb.set_trace()
image = feature_utils.all_to_gray(image)
image = image / 255.0
feature, descriptor = eng.vl_sift(
matlab.single(image.tolist()), 'Magnif', self.Magnif,
nargout=2) # peak_thresh=self.peak_thresh,
feature = np.transpose(np.array(feature))
descriptor = np.transpose(np.array(descriptor))
# pdb.set_trace()
return feature, descriptor
def extract_descriptor(self, image, feature):
commands = self._get_command_list('desc')
img = fu.all_to_gray(image)
cv2.imwrite(self.temp_img_path, img)
for command in commands:
subprocess.run(command, shell=True)
data_dict = loadh5(self.temp_desc_path)
desc = data_dict.get('descriptors')
return desc
def extract_descriptor(self, image, feature):
img = np.array(fu.all_to_gray(image), dtype='float32') / 255.0
pts, desc, _ = self.fe.run(img)
nb_kpts = MAX_CV_KPTS
#Already ordered by confidence
if pts.shape[1] < MAX_CV_KPTS:
nb_kpts = pts.shape[1]
if desc is not None:
desc = desc[:, :nb_kpts].T
return desc
def extract_descriptor(self, image, feature):
"""
feature = [[x, y, scale, angle]...]
"""
gray_image = utils.all_to_gray(image)
patches = []
for f in feature:
patch = utils.extract_patch(gray_image, f, patch_sz=32)
patches.append(patch)
patches = np.array(patches)
desc = self.extract_descriptors_from_patch_batch(patches)
return desc
def extract_all(self, image):
img = np.array(fu.all_to_gray(image), dtype='float32') / 255.0
start = time.time()
pts, desc, _ = self.fe.run(img)
nb_kpts = MAX_CV_KPTS
#Already ordered by confidence
if pts.shape[1] < MAX_CV_KPTS:
nb_kpts = pts.shape[1]
pts = pts[:2, :nb_kpts].T
if desc is not None:
desc = desc[:, :nb_kpts].T
return (pts, desc)
def detect_feature(self, image):
commands = self._get_command_list('kp')
img = fu.all_to_gray(image)
cv2.imwrite(self.temp_img_path, img)
for command in commands:
subprocess.run(command, shell=True)
kpts_raw = loadKpListFromTxt(self.temp_kpts_path)
kpts = []
for k in kpts_raw:
kpts.append([int(k[0]), int(k[1])])
kpts = np.array(kpts)
return kpts
def detect_feature(self, image):
image = fu.all_to_gray(image)
out = self.forward_passer(image)
kpts = out.ips_rc.T
print(kpts.shape)
order = np.argsort(out.ip_scores)[::-1]
num_kpts = MAX_KPTS
if len(order) < MAX_KPTS:
num_kpts = len(order)
order = order[:num_kpts]
kpts = kpts[order, :]
kpts = kpts[:, [1, 0]]
return kpts
def extract_all(self, image):
commands = self._get_command_list('desc')
img = fu.all_to_gray(image)
cv2.imwrite(self.temp_img_path, img)
for command in commands:
subprocess.run(command, shell=True)
kpts_raw = loadKpListFromTxt(self.temp_kpts_path)
data_dict = loadh5(self.temp_desc_path)
desc = data_dict.get('descriptors')
kpts = []
for k in kpts_raw:
kpts.append([int(k[0]), int(k[1])])
kpts = np.array(kpts)
return (kpts, desc)
def extract_descriptors_from_patch_batch(self, batch):
gray_batch = list()
for p in batch:
gray_batch.append(utils.all_to_gray(p))
batch = np.array(gray_batch)
batch = np.expand_dims(batch, axis=1)
#make multiple batches of Max batch size
splits = np.arange(0,batch.shape[0],MAX_BATCH)[1:]
batch_list = np.split(batch, splits)
descriptor_list = list()
for bat in batch_list:
if bat.size==0:
continue
bat = torch.from_numpy(bat)
descriptors = self.model(bat.float())
descriptor_list.extend(descriptors.detach().numpy().tolist())
descriptors = np.array(descriptor_list)
return descriptors
def extract_all(self, image):
img = fu.all_to_gray(image)
kpts, desc = self.run(img)
return (kpts, desc)
def extract_descriptor(self, image, feature):
img = fu.all_to_gray(image)
_, desc = self.run(img)
return desc
def detect_feature(self, image):
img = fu.all_to_gray(image)
kpts, _ = self.run(img)
return kpts
def detect_feature(self, image):
img = np.array(fu.all_to_gray(image), dtype='float32') / 255.0
pts, _, _ = self.fe.run(img)
pts = pts[:2, :].T
return pts
