def load_and_detect_images(im_path, tinyFaces_args, output_directory):
args = get_args(tinyFaces_args)
scaled_matrix = np.empty([])
results = []
print('-------Detecting images ... -------')
for files in tqdm(os.listdir(im_path)):
if files.endswith('.jpg') or files.endswith('.png'):
if os.path.isfile(output_directory + 'data/dataFrame_' +
files.split('.')[0] + '.pkl'):
print("-------> Loading DataFrame ...")
data_frame = read_pickle(output_directory + 'data/dataFrame_' +
files.split('.')[0] + '.pkl')
else:
print('-------> Imgage ' + files)
img = cv2.imread(os.getcwd() + '/' + im_path + files)
[scaled_matrix, bboxes, detected_faces_image
] = tinyFaces.tinyFaces_Detection(args, img)
index = []
rows = []
for i in range(len(scaled_matrix)):
rows.append({
'img': img,
'faces': scaled_matrix[i],
'bboxes': bboxes[i],
'name_img': files.split('.')[0]
})
index.append(i)
data_frame = DataFrame(rows, index=index)
data_frame.to_pickle(output_directory + 'data/dataFrame_' +
files.split('.')[0] + '.pkl')
# results.append([data_frame, img])
results.append(data_frame)
# data_frame --> shape (53,3)
return results
def main():
args = get_args()
depth = args.depth
k = args.width
max_age = args.max_age + 1
scaled_matrix = np.empty([])
for files in os.listdir(directory_files[0]):
if files.endswith('.jpg') or files.endswith('.png'):
print('-------Predicting image: ' + files + '-------')
# img = cv2.imread(os.getcwd()+'/'+directory_files[0]+files)
img = cv2.imread(os.getcwd() + '/' + im_path + files)
# for face detection
output_directory = directory_files[1] + files.split('.')[0]
if not os.path.exists(output_directory):
print("** Creating output_directory in " + output_directory +
' ... **')
os.makedirs(output_directory)
# [pnet, rnet, onet] = faceNet.create_FaceNet_network_Params(args)
# [scaled_matrix , n_faces_detected ,detected_faces_image] = faceNet.faceNet_Detection(img,output_directory, args, pnet, rnet, onet)
[scaled_matrix, n_faces_detected,
detected_faces_image] = tinyFaces.tinyFaces_Detection(args, img)
# Load model and weights of AGE-GENDER
img_size_age_gender = 64
img_size_ethnicity = 224
model_age_gender = WideResNet(img_size_age_gender,
depth=depth,
k=k,
units_age=max_age)()
model_age_gender.load_weights(weight_file)
#
# # Load model and weights of ETHNICITY
# model_ethnicity = create_face_network(nb_class=4, hidden_dim=512, shape=(224, 224, 3))
# model_ethnicity.load_weights(weights_ethnic_file)
# Resize the images for each model
faces = np.empty((len(scaled_matrix), img_size_age_gender,
img_size_age_gender, 3))
# faces1 = np.empty((len(scaled_matrix), img_size_ethnicity, img_size_ethnicity, 3))
for i in range(len(scaled_matrix)):
faces[i, :, :, :] = cv2.resize(
scaled_matrix[i],
(img_size_age_gender, img_size_age_gender))
# faces1[i, :, :, :] = cv2.resize(scaled_matrix[i], (img_size_ethnicity, img_size_ethnicity))
# # predict with ethnicity model
# # TODO : Reduce the number of predictions in a second / minute
# result_ethn = np.empty((len(faces1),4))
# for i in range(len(faces1)):
# result_ethn[i] = model_ethnicity.predict(transform_image_etnicity_to_predict(faces1[i]))
#NOTE predict ages and genders of the detected faces
result_age_gend = model_age_gender.predict(faces)
predicted_genders = result_age_gend[0]
ages = np.arange(0, max_age).reshape(max_age, 1)
predicted_ages = result_age_gend[1].dot(ages).flatten()
cols = 5
rows = int(len(scaled_matrix) / cols) + 1
for i in range(len(scaled_matrix)):
plt.subplot(rows, cols, i + 1)
im = faces[i].astype(np.uint8)
plt.imshow(cv2.cvtColor(im, cv2.COLOR_BGR2RGB))
plt.title("{}, {}".format(
int(predicted_ages[i]),
"F" if predicted_genders[i][0] > 0.5 else "M"))
plt.axis('off')
plt.subplots_adjust(hspace=0.6)
# cv2.imwrite(output_directory+'/'+"{}_{}_".format(int(predicted_ages[i]),
# "F" if predicted_genders[i][0] > 0.5 else "M") +"_id"+str(i)+".jpg", scaled_matrix[i])
plt.savefig(output_directory + "/result_" +
str(int(len(scaled_matrix))) + ".png")
def main():
args = get_args()
[pnet, rnet, onet] = faceNet.create_FaceNet_network_Params(args)
detected_faces = []
# for files in os.listdir(directory_files[0]):
# if files.endswith('.jpg') or files.endswith('.png'):
# print ('-------Analysing image: '+files+'-------')
# img = cv2.imread(os.getcwd()+'/'+directory_files[0]+files)
# # for face detection
# output_directory = directory_files[1]
# if not os.path.exists(output_directory):
# print ("** Creating output_directory in "+output_directory+' ... **')
# os.makedirs(output_directory)
#
#
# [scaled_matrix , n_faces_detected, detected_faces_image] = faceNet.faceNet_Detection(img,output_directory, args, pnet, rnet, onet)
# cv2.imwrite(output_directory+files,detected_faces_image)
#
# detected_faces.append(detected_faces_image)
#
# np.array(detected_faces).dump(open('detected_faces_array.npy', 'wb'))
for files in os.listdir(directory_files[0]):
if files.endswith('.jpg') or files.endswith('.png'):
print('-------Analysing image: ' + files + '-------')
img = cv2.imread(os.getcwd() + '/' + directory_files[0] + '/' +
files)
# for face detection
output_directory = directory_files[1]
if not os.path.exists(output_directory):
print("** Creating output_directory in " + output_directory +
' ... **')
os.makedirs(output_directory)
if args.face_detector == 'facenet':
# out_fn = cv2.VideoWriter('output_facenet.avi',fourcc, 20.0, size)
[scaled_matrix, n_faces_detected, detected_faces_image
] = faceNet.faceNet_Detection(img, output_directory, args,
pnet, rnet, onet)
# print ('scaled_matrix.shape',scaled_matrix.shape)
if detected_faces_image.ndim == 1:
print("--> SOMETHING IS WRONG IN THIS IMAGE")
else:
cv2.imwrite(output_directory + files, detected_faces_image)
detected_faces.append(scaled_matrix)
elif args.face_detector == 'tinyfaces':
# out_tf = cv2.VideoWriter('output_tinyfaces.avi',fourcc, 20.0, size)
[scaled_matrix, n_faces_detected, detected_faces_image
] = tinyFaces.tinyFaces_Detection(args, img)
# TODO: develop a scaled_matrix
cv2.imwrite(output_directory + files, detected_faces_image)
detected_faces.append(scaled_matrix)
else:
print(
'A face detector is required as an argument. The options are: facenet or tinyfaces.'
)
np.array(detected_faces).dump(
open('detected_faces_' + str(args.face_detector) + '_array.npy', 'wb'))
# detector = dlib.get_frontal_face_detector()
emotion_model = load_emotion_model('models/model.best.hdf5')
# [pnet, rnet, onet] = faceNet.create_FaceNet_network_Params(args)
try:
start = time.time()
im = cv.imread(im_path)
if not os.path.exists(output_directory):
print("** Creating output_directory in " + output_directory +
' ... **')
os.makedirs(output_directory)
# [scalied_matrix , n_faces_detected, detected_faces] = faceNet.faceNet_Detection(im,output_directory, args, pnet, rnet, onet)
[scaled_matrix, n_faces_detected,
detected_faces_image] = tinyFaces.tinyFaces_Detection(args, im)
# Resize the images for each model
faces = np.empty(
(len(scaled_matrix), img_size_emotions, img_size_emotions, 3))
cols = 5
rows = int(len(scaled_matrix) / cols) + 1
for i in range(len(scaled_matrix)):
faces[i, :, :, :] = cv.resize(
scaled_matrix[i], (img_size_emotions, img_size_emotions))
preds = emotion_model.predict(faces)
for i in range(len(scaled_matrix)):
def main():
start = time.time()
args = get_args()
depth = args.depth
k = args.width
capture = cv2.VideoCapture(video_directory)
# # NOTE: PYTHON 3
# fourcc = cv2.VideoWriter.fourcc('M','J','P','G')
# # In this way it always works, because your get the right "size"
# size = (int(capture.get(cv2.CAP_PROP_FRAME_WIDTH)),
# int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT)))
# NOTE: PYTHON 2.7
# fourcc = cv2.VideoWriter_fourcc('M','J','P','G')
fourcc = cv2.VideoWriter_fourcc('X', 'V', 'I', 'D')
size = (int(capture.get(cv2.CAP_PROP_FRAME_WIDTH)),
int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT)))
# out = cv2.VideoWriter('output.avi',fourcc, 20.0, size)
# out1 = cv2.VideoWriter('output_facenet.avi',fourcc, 20.0, size)
# out2 = cv2.VideoWriter('output_tinyfaces.avi',fourcc, 20.0, size)
if not os.path.exists(output_directory):
os.makedirs(output_directory)
if args.face_detector == 'facenet':
out1 = cv2.VideoWriter(output_directory + 'output_facenet.avi', fourcc,
20.0, size)
elif args.face_detector == 'tinyfaces':
out2 = cv2.VideoWriter(
os.path.join(output_directory, 'output_tinyfaces_TEST.avi'),
fourcc, 20.0, size)
count = 0
maxDetectedFaces = []
try:
while (capture.isOpened()):
success, frame = capture.read()
# print('Read a new frame: ', frame.shape) # --> (720, 1280, 3)
if args.face_detector == 'facenet':
# out_fn = cv2.VideoWriter('output_facenet.avi',fourcc, 20.0, size)
[pnet, rnet,
onet] = faceNet.create_FaceNet_network_Params(args)
[scaled_matrix, n_faces_detected, detected_faces_image
] = faceNet.faceNet_Detection(frame, output_directory, args,
pnet, rnet, onet)
# print ('scaled_matrix.shape',scaled_matrix.shape)
# if detected_faces_image.ndim == 1:
# print("------SOMETHING IS WRONG------")
# sys.exit(0)
# else:
# IDEA: LOAD MODELS -- SLOW
# NOTE[faces, faces1] = loadModels(scaled_matrix, depth, k)
# FACES SHAPEEEE.------ (11, 64, 64, 3)
# FACES 11111 SHAPEEEE.------ (11, 224, 224, 3)
# cv2.imshow('FRAME',detected_faces_image)
# IDEA: save all frames
# name = "out_video/frame_"+str(count)+".jpg"
# print (name)
# cv2.imwrite(name,detected_faces_image)
out1.write(detected_faces_image)
count += 1
elif args.face_detector == 'tinyfaces':
# out_tf = cv2.VideoWriter('output_tinyfaces.avi',fourcc, 20.0, size)
try:
[scaled_matrix, n_faces_detected, detected_faces_image
] = tinyFaces.tinyFaces_Detection(args, frame)
# TODO: develop a scaled_matrix
# if detected_faces_image.ndim == 1:
# print("------SOMETHING IS WRONG------")
# else:
out2.write(detected_faces_image)
maxDetectedFaces.append(len(n_faces_detected))
cv2.imwrite(
output_directory +
'output_video_tinyFaces/frame_%05d.png' % count,
detected_faces_image)
count += 1
except:
print("This is an error message in frame " + str(count))
break
else:
print(
'A face detector is required as an argument. The options are: facenet or tinyfaces.'
)
text_file = open("MAX_COUNTER.txt", "w")
text_file.write("MAXIM COUNTER: %s" % np.max(maxDetectedFaces))
text_file.close()
print("Total time: ", time.time() - start)
# Release everything if job is finished
capture.release()
# out1.release()
# out2.release()
cv2.destroyAllWindows()
except KeyboardInterrupt:
aux = time.time() - start
capture.release()
out1.release()
out2.release()
print("Total time: ", aux)
cv2.destroyAllWindows()