def main(image_file):
"""
Get a uploaded file, crop faces, assign to result
:param image_file: input file
:return: list of human faces and their hyperparameters and original image in base64 format to display
"""
img_input = np.fromstring(image_file.read(),
np.uint8) # read file to input stream
img = cv2.imdecode(img_input,
cv2.IMREAD_COLOR) # decode input stream to numpy array
original_img_base64 = predict.nparray_to_base64(
img) # return original image with bas64 format
results = []
detected_faces = predict.face_detector(img) # detect faces in image
if detected_faces:
for i, face in enumerate(detected_faces):
cropped_face = predict.nparray_to_base64(
face) # export to tag <img> in html
prob_age = predict.predict_age(face) # call model to predict age
age_data = [math.floor(100 * x) for x in prob_age[0]
] # change probabilities to percentages
grouped_age = np.argmax(
prob_age[0]) # get the highest probability in prediction
if grouped_age == 0: # split age in range
displayed_age = "0 ~ 10 years old"
else:
displayed_age = str(grouped_age) + "0" + " ~ " + str(
grouped_age) + "9" + " years old"
prob_gender = predict.predict_gender(
face) # call model to predict gender
if prob_gender > 0.5:
displayed_gender = "Male"
else:
displayed_gender = "Female"
male_prob = math.floor(100 *
prob_gender[0]) # change to percentages
female_prob = 100 - male_prob
results.append({
'index_gender': 'canvas_gender_' + str(i),
'index_age': 'canvas_age_' + str(i),
'cropped_image': cropped_face,
'target_gender': displayed_gender,
'target_age': displayed_age,
'prob_male': male_prob,
'prob_female': female_prob,
'prob_age': age_data,
})
else:
print("Error: cannot detect any human faces in the picture")
return original_img_base64, results
def main_stream():
"""
Streaming results to image tag
:param :
:return: flow of video frame
"""
# ID verification
previous_id = np.zeros(shape=(1, 128))
detected_id_threshold = -0.9
id_count = 0
frame_rate = 25 # adjust frame rate from camera
prev = 0
alpha = 1.5 # border of face
while True:
time_elapsed = time.time() - prev #
ret, image = cap.read() # get video frame
if time_elapsed > 1. / frame_rate:
prev = time.time()
_, detections = predict.face_detector(
image) # detect face in a picture
detected, crop_face, top_left, bottom_right = nearest_standing(
image, detections, alpha)
content = ""
if detected:
vector_face = face_recognition.face_encodings(crop_face)
if vector_face:
age_prob = predict.predict_age_id(crop_face)
gender_prob = predict.predict_gender(crop_face)
text_gender = "M" if gender_prob[0][0] > 0.5 else "F"
text_age = str(np.argmax(age_prob[0]))
detected_id = tf.keras.losses.cosine_similarity(
previous_id, vector_face[0]).numpy()
if detected_id > detected_id_threshold:
id_count += 1
previous_id = vector_face[0]
content = "G: " + text_gender + ", R: " + text_age + ", ID: " + str(
id_count)
image = draw_label(image, top_left, bottom_right, content)
if not ret:
print("Error: failed to capture image")
break
yield (b'--frame\r\n'
b'Content-Type: image/jpeg\r\n\r\n' +
cv2.imencode('.jpg', image)[1].tostring() + b'\r\n')
def generate_api_stream():
"""
Streaming results as json
:param :
:return: flow of video frame as json
"""
# ID verification
previous_id = np.zeros(shape=(1, 128))
detected_id_threshold = -0.9
id_count = 0
frame_rate = 25 # adjust framerate from camera
prev = 0
alpha = 1.5
while True:
time_elapsed = time.time() - prev #
ret, image = cap.read() # get video frame
if time_elapsed > 1. / frame_rate:
prev = time.time()
_, detections = predict.face_detector(image) # detect face in a picture
detected, crop_face, top_left, bottom_right = streaming.nearest_standing(image, detections, alpha)
content = {}
if detected:
vector_face = face_recognition.face_encodings(crop_face)
if vector_face:
age_prob = predict.predict_age_id(crop_face)
gender_prob = predict.predict_gender(crop_face)
text_gender = "M" if gender_prob[0][0] > 0.5 else "F"
text_age = str(np.argmax(age_prob[0]))
detected_id = tf.keras.losses.cosine_similarity(previous_id, vector_face[0]).numpy()
if detected_id > detected_id_threshold:
id_count += 1
previous_id = vector_face[0]
# content = "G: " + text_gender + ", R: " + text_age + ", ID: " + str(id_count)
content = {
'gender': text_gender,
'age': text_age,
'id': identification,
'time': datetime.datetime.now().strftime("%Y/%m/%d, %H:%M:%S"),
'vector': vector_face[0].tolist(),
}
if not ret:
print("Error: failed to capture image")
break
yield json.dumps(content)
def capture_frame():
"""
Capture one frame and predict
:param :
:return: json file
"""
global identification, previous_id
# ID verification
detected_id_threshold = -0.9
frame_rate = 25 # adjust framerate from camera
prev = 0
alpha = 1.5
ret, image = cap.read() # get video frame
_, detections = predict.face_detector(image) # detect face in a picture
detected, crop_face, top_left, bottom_right = streaming.nearest_standing(image, detections, alpha)
content = {}
if detected:
vector_face = face_recognition.face_encodings(crop_face)
if vector_face:
age_prob = predict.predict_age_id(crop_face)
gender_prob = predict.predict_gender(crop_face)
text_gender = "M" if gender_prob[0][0] > 0.5 else "F"
text_age = str(np.argmax(age_prob[0]))
detected_id = tf.keras.losses.cosine_similarity(previous_id, vector_face[0]).numpy()
if detected_id > detected_id_threshold:
identification += 1
previous_id = vector_face[0]
# content = "G: " + text_gender + ", R: " + text_age + ", ID: " + str(identification)
content = {
'gender': text_gender,
'age': text_age,
'id': identification,
'time': datetime.datetime.now(),
'vector': vector_face[0].tolist(),
}
else:
pass # content is a blank dictionary
print(jsonify(content))
return jsonify(content)
def generate_api_stream():
"""
Streaming results as json
:param :
:return: flow of video frame as json and write record to database
"""
# set up database
conn, path_to_db = setup_database() # create new database sqlite
record_count = 0 # counter of record is limited in one database
record_max = 100000 # maximum of records in a table
# ID verification
previous_id = np.zeros(shape=(1, 128))
detected_id_threshold = -0.9
id_count = 0
frame_rate = 25 # adjust framerate from camera
prev = 0 # time counter
alpha = 1.5 # border of faces
while True:
time_elapsed = time.time() - prev
ret, image = cap.read() # get video frame
if time_elapsed > 1. / frame_rate:
prev = time.time()
record_count += 1 # record is written
if record_count == record_max: # reach limitation
conn, path_to_db = setup_database() # setup new database
_, detections = predict.face_detector(
image) # detect face in a picture
detected, crop_face, top_left, bottom_right = streaming.nearest_standing(
image, detections, alpha)
content = {}
if detected:
vector_face = face_recognition.face_encodings(crop_face)
if vector_face:
age_prob = predict.predict_age_id(crop_face)
gender_prob = predict.predict_gender(crop_face)
text_gender = "M" if gender_prob[0][0] > 0.5 else "F"
text_age = str(np.argmax(age_prob[0]))
detected_id = tf.keras.losses.cosine_similarity(
previous_id, vector_face[0]).numpy()
if detected_id > detected_id_threshold:
id_count += 1
previous_id = vector_face[0]
# save content into dictionary
content = {
'gender': text_gender,
'age': text_age,
'id': id_count,
'vector': json.dumps(vector_face[0].tolist()),
}
# write record to database
create_record(conn, (content['gender'], content['age'],
content['id'], content['vector']))
if not ret:
print("Error: failed to capture image")
break
yield json.dumps(
content) # streaming dictionary to front end