def Train_Database_Sign(self,
database,
version,
num_of_kpts,
force_update = False,
K = 10,#tree branch
depth = 4, #depth
nleaves = 10000 #leaves in the tree
):
#load information from database
num_in_set = database[0]
num_of_sets = database[1]
data_dir = database[2]
total = num_in_set * num_of_sets
updated = force_update # will be turned true if one of the steps has been processed, so that the following step will be focused to processs!
'''
generate desc~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ dir should be saved
'''
desc_dir = os.path.join(self.DESC_DIR, 'db_version_'+str(version), str(num_of_kpts))
if updated or (not os.path.isdir(desc_dir) or os.listdir(desc_dir) == []):
updated = True
if not os.path.isdir(os.path.join(self.DESC_DIR, 'db_version_'+str(version))):
os.mkdir(os.path.join(self.DESC_DIR, 'db_version_'+str(version)))
if not os.path.isdir(desc_dir):
os.mkdir(desc_dir)
for d in data_dir:
img_idx = d[0]
class_idx = d[1]
img_dir = d[2]
'''
k is the index of desc included in the name, should be ordered by the class
'''
k = img_idx + class_idx * num_in_set
#load image(load image, convert to np array with float type)
img = Image.open(img_dir).convert('L')
img_s = self.StandalizeImage(img, 480)
img_data = np.asarray(img_s, dtype=float)
kp = Keypoint()
kp.generate_keypoint(num_of_kpts, img.size[0], img.size[1], self.SIGMA) #sigma is set to 1 currently
#kp.save_keypoint(self.KEYPOINT_DIR, self.KEYPOINT_FILE+str(num_of_kpts)) #save to a directory
print 'Random keypoint generated'
#generate desc
desc = Descriptor()
desc.generate_desc(img_data, kp.kpt)
desc.save_desc(desc_dir, self.DESC_FILE + str(k))
print desc.desc
#load to a large matrix
#desc_database[:, k*num_of_kpts:k+num_of_kpts] = desc.desc #add to the database therefore later can be used to train the tree
print '=>'+str(k) ,
print 'Descriptor Generated'
#load desc
desc_database = np.empty(shape=(128, total*num_of_kpts), dtype=np.uint8)
for k in range(0, total):
desc = Descriptor()
desc.load_desc(desc_dir, self.DESC_FILE + str(k))
desc_database[:, k*num_of_kpts:(k+1)*num_of_kpts] = desc.desc
print 'Descriptor Loaded'
'''
Build the tree~~~~~~~~~~~~~~~~~~~~~~~~~
'''
tree_dir = os.path.join(self.TREE_DIR, 'db_version_' + str(version))
if updated or (not os.path.isfile(os.path.join(tree_dir, str(num_of_kpts) + self.TREE_FILE))):
updated = True
if not os.path.isdir(tree_dir):
os.mkdir(tree_dir)
tree = Tree()
tree.generate_tree(desc_database, K, nleaves, tree_dir, str(num_of_kpts) + self.TREE_FILE)
print 'Tree built'
'''
Generate signature~~~~~~~~~~~~~~~~~~~~~~~~~
'''
sign_dir = os.path.join(self.SIGN_DIR, 'db_version_' + str(version))
if updated or (not os.path.isfile(os.path.join(sign_dir, self.SIGN_FILE+str(num_of_kpts)))):
updated = True
tr = vl._vlfeat.VlHIKMTree(0, 0)
tr.load(os.path.join(tree_dir, str(num_of_kpts) + self.TREE_FILE))
print 'Tree Loaded'
sign = Signature()
sign.generate_sign_database_dir(tr, desc_database, K, depth, total, num_of_kpts)
if not os.path.isdir(sign_dir):
os.mkdir(sign_dir)
sign.save_sign(sign_dir, self.SIGN_FILE+str(num_of_kpts))
print 'Signature Generated'
else:
print 'Signature Already Generated'
del desc_database
return updated;
# x = np.linspace(0, 1, 200) # y = x ** 2 x = np.linspace(0, 2 * np.pi, 100) y = np.sin(x) noise = np.random.rand(y.shape[0]) * 0.4 y = y + noise data = np.array([x, y]).T fig_data = plt.figure(figsize=(20, 10)) plt.scatter(data[:, 0], data[:, 1]) plt.show() tree = Tree(data, max_depth=7) tree.generate_tree(error=5.0e-2) x_max = x.max() x_min = x.min() y_max = y.max() y_min = y.min() x_v = np.linspace(x_min, x_max, 500) y_v = np.linspace(y_min, y_max, 100) x_im, y_im = np.meshgrid(x_v, y_v) z = np.zeros(x_im.shape) count = x_im.shape[0] * x_im.shape[1] pos = 0 for i in np.arange(0, x_im.shape[0], 1): for j in np.arange(0, x_im.shape[1], 1):
