def get_topic_proportions_for_every_image():
from dir_processing import DirProcessing
landmarks_urls_list = []
person_ids = DirProcessing.get_all_person_ids()
for person_id in person_ids:
perform_ids = DirProcessing.get_all_perform_ids_from_person_id(person_id)
for perform_id in perform_ids:
landmarks_urls = DirProcessing.get_all_landmarks_urls_from_sequence(person_id, perform_id)
landmarks_urls_list.extend(landmarks_urls)
doc_num = len(landmarks_urls_list)
dt_file = '../ctm-dist/CTM46/final-lambda.dat'
dt_vector = np.loadtxt(dt_file)
topic_num = dt_vector.size / doc_num
dt_matrix = np.reshape(dt_vector, (doc_num, topic_num))
np.set_printoptions(suppress=True)
final_theta = np.exp(dt_matrix)
final_theta = final_theta / np.sum(final_theta, axis=1)[:, np.newaxis]
return landmarks_urls_list, final_theta
def generate_features_pool(self):
""" generate train and test files for classification """
from analysis import Analysis
from dir_processing import DirProcessing
from file_op import FileOp
landmarks_urls_list, features = Analysis.get_topic_proportions_for_every_image()
subsets_dict = self.divide_persons_into_subsets()
for i in range(0, len(landmarks_urls_list)):
landmarks_url = landmarks_urls_list[i]
label_url = DirProcessing.get_label_url_from_landmarks_url(landmarks_url)
loc = DirProcessing.get_location_from_sequence(landmarks_url, 3)
if label_url and loc != "MIDDLE":
person_id, perform_id, index_id = DirProcessing.get_id_from_label_url(label_url)
subset_id = subsets_dict[person_id]
feature = features[i, :]
if loc == "START":
label = 0
else:
label = FileOp.read_label_file(label_url)
self.features_pool[subset_id].append(feature)
self.labels_pool[subset_id].append(label)
self.urls_pool[subset_id].append(landmarks_url)
print "Features pools have been generated. "
def test():
import os
import sys
lib_path = os.path.abspath('../utilization/')
sys.path.append(lib_path)
from dir_processing import DirProcessing
from file_op import FileOp
LSF.build_dictionary()
landmarks_urls = DirProcessing.get_all_landmarks_urls_from_sequence('87','4')
test_one_sequence = True
if test_one_sequence:
video = LSF.lsf_from_sequence(landmarks_urls)
print video.landmarks_urls
test_one_landmark = False
if test_one_landmark:
landmarks = FileOp.read_landmarks_file(landmarks_urls[11])
landmarks_neural = FileOp.read_landmarks_file(landmarks_urls[0])
lsf_document = LSF.lsf(landmarks, landmarks_neural)
landmark = landmarks[35,:]
landmark_neural = landmarks_neural[35, :]
word = LSF.compute_word(35, landmark, landmark_neural)
print landmark, landmark_neural, word
def draw_landmarks_on_sequence(img_urls):
from dir_processing import DirProcessing
landmarks_urls = []
for i in xrange(0, len(img_urls)):
img_url = img_urls[i]
landmarks_url = DirProcessing.generate_landmarks_url_from_img_url(img_url)
landmarks_urls.append(landmarks_url)
from file_op import FileOp
for i in xrange(0, len(img_urls)):
img_url = img_urls[i]
landmarks_url = landmarks_urls[i]
img = FileOp.read_img_file(img_url)
landmarks = FileOp.read_landmarks_file(landmarks_url)
for i in xrange(0, landmarks.shape[0]): # for every point
loc = landmarks[i, :]
x = int(round(loc[0]))
y = int(round(loc[1]))
cv2.circle(img, (x, y), 1, 255)
print img_url
cv2.imshow('image', img)
k = cv2.waitKey(0)
if k == 27: # Esc key to stop
break
cv2.destroyAllWindows()
def test():
import os, sys
lib_path = os.path.abspath('../utilization/')
sys.path.append(lib_path)
from dir_processing import DirProcessing
label_url = DirProcessing.generate_label_url('34','3','27')
label = FileOp.read_label_file(label_url)
print label
def test():
lib_path = os.path.abspath('../utilization/')
sys.path.append(lib_path)
from dir_processing import DirProcessing
img_urls = DirProcessing.get_all_img_urls_from_sequence('109','2')
print img_urls
Visualization.draw_landmarks_on_sequence(img_urls)
def generate_corpus_and_write_to_file():
""" generate the copus, write it to files and store the LSF corpus features """
import os
import sys
lib_path = os.path.abspath('../utilization/')
sys.path.append(lib_path)
from dir_processing import DirProcessing
LSF.build_dictionary()
lsf_corpus = []
person_ids = DirProcessing.get_all_person_ids()
for person_id in person_ids:
perform_ids = DirProcessing.get_all_perform_ids_from_person_id(person_id)
for perform_id in perform_ids:
landmarks_urls = DirProcessing.get_all_landmarks_urls_from_sequence(person_id, perform_id)
expression_sequence = LSF.lsf_from_sequence(landmarks_urls)
print 'The feature extraction of expression person S{} and perform time {} has ' \
'been done.'.format(person_id, perform_id)
lsf_corpus.append(expression_sequence)
import cPickle
with open('../model/corpus.pk', 'wb') as f:
cPickle.dump(lsf_corpus, f)
with open('../model/corpus.txt', 'w') as f:
for expression_sequence in lsf_corpus:
lsf_sequence = expression_sequence.lsf_sequence
for lsf_document in lsf_sequence:
f.write(str(len(lsf_document)))
for word, count in lsf_document.iteritems():
wid = LSF.word2id[word]
s = " %d:%d" %(wid, count)
f.write(s)
f.write("\n")
def demonstrate_wrong_predictions(self):
from dir_processing import DirProcessing
from visualization import Visualization
for i in range(len(self.predict_labels)):
test_label = self.test_labels[i]
predict_label = self.predict_labels[i]
if test_label != predict_label:
test_expression = Classifier.get_expression_from_label(test_label)
predict_expression = Classifier.get_expression_from_label(predict_label)
print "Wrong prediction: {}, Right prediction: {}.".format(predict_expression, test_expression)
landmarks_url = self.test_urls[i]
img_url = DirProcessing.generate_img_url_from_landmarks_url(landmarks_url)
Visualization.draw_landmarks_on_img(img_url)
def random_show_cluster(landmarks_urls_list, label, cluster_index):
from dir_processing import DirProcessing
from file_op import FileOp
from visualization import Visualization
sample_list = np.where(label == cluster_index)[0]
sample_list = np.random.permutation(sample_list)
print len(sample_list)
landmarks_urls_sublist = []
for i in sample_list:
landmarks_urls_sublist.append(landmarks_urls_list[i])
img_urls_sublist = DirProcessing.generate_img_urls_from_landmarks_urls(landmarks_urls_sublist)
Visualization.draw_landmarks_on_sequence(img_urls_sublist)
def divide_persons_into_subsets(self):
from dir_processing import DirProcessing
person_ids = DirProcessing.get_all_person_ids()
permute_ids = np.random.permutation(person_ids)
total_person_num = len(person_ids)
step_arr = np.linspace(0, total_person_num, self.subset_num + 1)
step_arr = step_arr.astype(int)
subsets_dict = dict.fromkeys(person_ids)
for i in range(0, self.subset_num):
for j in range(step_arr[i], step_arr[i + 1]):
subsets_dict[permute_ids[j]] = i
return subsets_dict
def draw_landmarks_on_img(img_url, drawtext=False):
from dir_processing import DirProcessing
landmarks_url = DirProcessing.generate_landmarks_url_from_img_url(img_url)
from file_op import FileOp
img = FileOp.read_img_file(img_url)
landmarks = FileOp.read_landmarks_file(landmarks_url)
for i in xrange(0, landmarks.shape[0]): # for every point
loc = landmarks[i, :]
x = int(round(loc[0]))
y = int(round(loc[1]))
cv2.circle(img, (x, y), 2, 255)
if drawtext:
cv2.putText(img, str(i), (x,y), cv2.FONT_HERSHEY_SIMPLEX, 0.25, 255)
cv2.imshow('image',img)
cv2.waitKey(0)
cv2.destroyAllWindows()