def load_model(weights = AGE_WEIGHTS):
stage_num = [3,3,3]
lambda_local = 1
lambda_d = 1
model = SSR_net(IMG_SIZE,stage_num, lambda_local, lambda_d)()
model.load_weights(weights)
return model
def model_init(args):
K.set_learning_phase(0) # make sure its testing mode
# load model and weights
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model_age = SSR_net(args.image_size, stage_num, lambda_local, lambda_d)()
model_age.load_weights(args.age_model)
model_gender = SSR_net_general(args.image_size, stage_num, lambda_local,
lambda_d)()
model_gender.load_weights(args.gender_model)
return model_gender, model_age
def main():
weight_file = "./ssrnet_3_3_3_64_1.0_1.0.h5"
# load model and weights
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model.save('ssrnet.h5')
model.summary()
coreml_model = coremltools.converters.keras.convert(
model,
input_names="image",
image_input_names="image",
output_names="output",
add_custom_layers=True,
custom_conversion_functions={"Lambda": convert_lambda})
# Look at the layers in the converted Core ML model.
print("\nLayers in the converted model:")
for i, layer in enumerate(coreml_model._spec.neuralNetwork.layers):
if layer.HasField("custom"):
print("Layer %d = %s --> custom layer = %s" %
(i, layer.name, layer.custom.className))
else:
print("Layer %d = %s" % (i, layer.name))
coreml_model.save('ssrnet.mlmodel')
def load_models():
weight_file = os.path.join(
ssrnet_dir,
'../pre-trained/morph2/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5'
)
weight_file_gender = os.path.join(
ssrnet_dir,
'../pre-trained/wiki_gender_models/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5'
)
# load model and weights
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model_gender = SSR_net_general(img_size, stage_num, lambda_local,
lambda_d)()
model_gender.load_weights(weight_file_gender)
return model, model_gender
def main():
logger.info('Load InceptionV3 model')
inceptionv3 = AgenderNetInceptionV3()
inceptionv3.setWeight('trainweight/inceptionv3_2/model.16-3.7887-0.9004-6.6744.h5')
logger.info('Load MobileNetV2 model')
mobilenetv2 = AgenderNetMobileNetV2()
mobilenetv2.setWeight('trainweight/mobilenetv2/model.10-3.8290-0.8965-6.9498.h5')
logger.info('Load SSRNet model')
ssrnet = SSRNet(64, [3, 3, 3], 1.0, 1.0)
ssrnet.setWeight('trainweight/ssrnet/model.37-7.3318-0.8643-7.1952.h5')
logger.info('Load pretrain imdb model')
imdb_model = SSR_net(64, [3, 3, 3], 1.0, 1.0)()
imdb_model.load_weights("tes_ssrnet/imdb_age_ssrnet_3_3_3_64_1.0_1.0.h5")
imdb_model_gender = SSR_net_general(64, [3, 3, 3], 1.0, 1.0)()
imdb_model_gender.load_weights("tes_ssrnet/imdb_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
logger.info('Load pretrain wiki model')
wiki_model = SSR_net(64, [3, 3, 3], 1.0, 1.0)()
wiki_model.load_weights("tes_ssrnet/wiki_age_ssrnet_3_3_3_64_1.0_1.0.h5")
wiki_model_gender = SSR_net_general(64, [3, 3, 3], 1.0, 1.0)()
wiki_model_gender.load_weights("tes_ssrnet/wiki_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
logger.info('Load pretrain morph model')
morph_model = SSR_net(64, [3, 3, 3], 1.0, 1.0)()
morph_model.load_weights("tes_ssrnet/morph_age_ssrnet_3_3_3_64_1.0_1.0.h5")
morph_model_gender = SSR_net_general(64, [3, 3, 3], 1.0, 1.0)()
morph_model_gender.load_weights("tes_ssrnet/morph_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
utk = pd.read_csv('dataset/UTKface.csv')
fgnet = pd.read_csv('dataset/FGNET.csv')
utk_paths = utk['full_path'].values
fgnet_paths = fgnet['full_path'].values
logger.info('Read UTKface aligned images')
utk_images = [cv2.imread('UTKface_aligned/'+path) for path in tqdm(utk_paths)]
logger.info('Read FGNET aligned images')
fgnet_images = [cv2.imread('FGNET_aligned/'+path) for path in tqdm(fgnet_paths)]
utk_X = np.array(utk_images)
fgnet_X = np.array(fgnet_images)
utk_pred_age = dict()
utk_pred_gender = dict()
fgnet_pred_age = dict()
logger.info('Predict with InceptionV3')
start = time.time()
utk_pred_gender['inceptionv3'], utk_pred_age['inceptionv3'] = get_result(inceptionv3, utk_X)
_, fgnet_pred_age['inceptionv3'] = get_result(inceptionv3, fgnet_X)
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
del utk_X, fgnet_X
logger.info('Resize image to 96 for MobileNetV2')
utk_images = [cv2.resize(image, (96, 96), interpolation = cv2.INTER_CUBIC) for image in tqdm(utk_images)]
fgnet_images = [cv2.resize(image, (96, 96), interpolation = cv2.INTER_CUBIC) for image in tqdm(fgnet_images)]
utk_X = np.array(utk_images)
fgnet_X = np.array(fgnet_images)
logger.info('Predict with MobileNetV2')
start = time.time()
utk_pred_gender['mobilenetv2'], utk_pred_age['mobilenetv2'] = get_result(mobilenetv2, utk_X)
_, fgnet_pred_age['mobilenetv2'] = get_result(mobilenetv2, fgnet_X)
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
del utk_X, fgnet_X
logger.info('Resize image to 64 for SSR-Net')
utk_images = [cv2.resize(image, (64, 64), interpolation = cv2.INTER_CUBIC) for image in tqdm(utk_images)]
fgnet_images = [cv2.resize(image, (64, 64), interpolation = cv2.INTER_CUBIC) for image in tqdm(fgnet_images)]
utk_X = np.array(utk_images)
fgnet_X = np.array(fgnet_images)
logger.info('Predict with SSR-Net')
start = time.time()
utk_pred_gender['ssrnet'], utk_pred_age['ssrnet'] = get_result(ssrnet, utk_X)
_, fgnet_pred_age['ssrnet'] = get_result(ssrnet, fgnet_X)
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
logger.info('Predict with IMDB_SSR-Net')
start = time.time()
utk_pred_gender['ssrnet-imdb'] = np.around(imdb_model_gender.predict(utk_X).squeeze()).astype('int')
utk_pred_age['ssrnet-imdb'] = imdb_model.predict(utk_X).squeeze()
fgnet_pred_age['ssrnet-imdb'] = imdb_model.predict(fgnet_X).squeeze()
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
logger.info('Predict with Wiki_SSR-Net')
start = time.time()
utk_pred_gender['ssrnet-wiki'] = np.around(wiki_model_gender.predict(utk_X).squeeze()).astype('int')
utk_pred_age['ssrnet-wiki'] = wiki_model.predict(utk_X).squeeze()
fgnet_pred_age['ssrnet-wiki'] = wiki_model.predict(fgnet_X).squeeze()
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
logger.info('Predict with Morph_SSR-Net')
start = time.time()
utk_pred_gender['ssrnet-morph'] = np.around(morph_model_gender.predict(utk_X).squeeze()).astype('int')
utk_pred_age['ssrnet-morph'] = morph_model.predict(utk_X).squeeze()
fgnet_pred_age['ssrnet-morph'] = morph_model.predict(fgnet_X).squeeze()
elapsed = time.time() - start
logger.info('Time elapsed {:.2f} sec'.format(elapsed))
utk_pred_age = pd.DataFrame.from_dict(utk_pred_age)
utk_pred_gender = pd.DataFrame.from_dict(utk_pred_gender)
fgnet_pred_age = pd.DataFrame.from_dict(fgnet_pred_age)
utk_pred_age = pd.concat([utk['age'], utk_pred_age], axis=1)
utk_pred_gender = pd.concat([utk['gender'], utk_pred_gender], axis=1)
fgnet_pred_age = pd.concat([fgnet['age'], fgnet_pred_age], axis=1)
utk_pred_age.to_csv('result/utk_age_prediction.csv', index=False)
utk_pred_age.to_csv('result/utk_gender_prediction.csv', index=False)
fgnet_pred_age.to_csv('result/fgnet_age_prediction.csv', index=False)
def main():
K.set_learning_phase(0) # make sure its testing mode
weight_file = "../pre-trained/wiki/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
# for face detection
# detector = dlib.get_frontal_face_detector()
detector = MTCNN()
try:
os.mkdir('./img')
except OSError:
pass
# load model and weights
img_size = 64
stage_num = [3,3,3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size,stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
clip = VideoFileClip(sys.argv[1]) # can be gif or movie
#python version
pyFlag = ''
if len(sys.argv)<3 :
pyFlag = '2' #default to use moviepy to show, this can work on python2.7 and python3.5
elif len(sys.argv)==3:
pyFlag = sys.argv[2] #python version
else:
print('Wrong input!')
sys.exit()
img_idx = 0
detected = '' #make this not local variable
time_detection = 0
time_network = 0
time_plot = 0
ad = 0.4
skip_frame = 5 # every 5 frame do 1 detection and network forward propagation
for img in clip.iter_frames():
img_idx = img_idx + 1
input_img = img #using python2.7 with moivepy to show th image without channel flip
if pyFlag == '3':
input_img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_h, img_w, _ = np.shape(input_img)
input_img = cv2.resize(input_img, (1024, int(1024*img_h/img_w)))
img_h, img_w, _ = np.shape(input_img)
if img_idx==1 or img_idx%skip_frame == 0:
# detect faces using dlib detector
start_time = timeit.default_timer()
detected = detector.detect_faces(input_img)
elapsed_time = timeit.default_timer()-start_time
time_detection = time_detection + elapsed_time
faces = np.empty((len(detected), img_size, img_size, 3))
for i, d in enumerate(detected):
print(i)
print(d['confidence'])
if d['confidence'] > 0.95:
x1,y1,w,h = d['box']
x2 = x1 + w
y2 = y1 + h
xw1 = max(int(x1 - ad * w), 0)
yw1 = max(int(y1 - ad * h), 0)
xw2 = min(int(x2 + ad * w), img_w - 1)
yw2 = min(int(y2 + ad * h), img_h - 1)
cv2.rectangle(input_img, (x1, y1), (x2, y2), (255, 0, 0), 2)
# cv2.rectangle(img, (xw1, yw1), (xw2, yw2), (255, 0, 0), 2)
faces[i,:,:,:] = cv2.resize(input_img[yw1:yw2 + 1, xw1:xw2 + 1, :], (img_size, img_size))
start_time = timeit.default_timer()
if len(detected) > 0:
# predict ages and genders of the detected faces
results = model.predict(faces)
predicted_ages = results
# draw results
for i, d in enumerate(detected):
if d['confidence'] > 0.95:
x1,y1,w,h = d['box']
label = "{}".format(int(predicted_ages[i]))
draw_label(input_img, (x1, y1), label)
elapsed_time = timeit.default_timer()-start_time
time_network = time_network + elapsed_time
start_time = timeit.default_timer()
if pyFlag == '2':
img_clip = ImageClip(input_img)
img_clip.show()
cv2.imwrite('img/'+str(img_idx)+'.png',cv2.cvtColor(input_img, cv2.COLOR_RGB2BGR))
elif pyFlag == '3':
cv2.imshow("result", input_img)
cv2.imwrite('img/'+str(img_idx)+'.png',cv2.cvtColor(input_img, cv2.COLOR_RGB2BGR))
elapsed_time = timeit.default_timer()-start_time
time_plot = time_plot + elapsed_time
else:
for i, d in enumerate(detected):
if d['confidence'] > 0.95:
x1,y1,w,h = d['box']
x2 = x1 + w
y2 = y1 + h
xw1 = max(int(x1 - ad * w), 0)
yw1 = max(int(y1 - ad * h), 0)
xw2 = min(int(x2 + ad * w), img_w - 1)
yw2 = min(int(y2 + ad * h), img_h - 1)
cv2.rectangle(input_img, (x1, y1), (x2, y2), (255, 0, 0), 2)
# cv2.rectangle(img, (xw1, yw1), (xw2, yw2), (255, 0, 0), 2)
faces[i,:,:,:] = cv2.resize(input_img[yw1:yw2 + 1, xw1:xw2 + 1, :], (img_size, img_size))
# draw results
for i, d in enumerate(detected):
if d['confidence'] > 0.95:
x1,y1,w,h = d['box']
label = "{}".format(int(predicted_ages[i]))
draw_label(input_img, (x1, y1), label)
start_time = timeit.default_timer()
if pyFlag == '2':
img_clip = ImageClip(input_img)
img_clip.show()
elif pyFlag == '3':
cv2.imshow("result", input_img)
elapsed_time = timeit.default_timer()-start_time
time_plot = time_plot + elapsed_time
#Show the time cost (fps)
print('avefps_time_detection:',img_idx/time_detection)
print('avefps_time_network:',img_idx/time_network)
print('avefps_time_plot:',img_idx/time_plot)
print('===============================')
if pyFlag == '3':
key = cv2.waitKey(30)
if key == 27:
break
async def main():
weight_file = "../pre-trained/megaface_asian/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
weight_file_gender = "../pre-trained/wiki_gender_models/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
mtccn = False
if mtccn:
detector = MTCNN()
else:
detector = cv2.CascadeClassifier('lbpcascade_frontalface_improved.xml')
# load model and weights
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model_gender = SSR_net_general(img_size, stage_num, lambda_local,
lambda_d)()
model_gender.load_weights(weight_file_gender)
# capture video
cap = cv2.VideoCapture(0)
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1024 * 1)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 768 * 1)
detected = ''
time_detection = 0
time_network = 0
time_plot = 0
ad = 0.5
img_idx = 0
skip_frame = 10
skip_frame = 10
sleep = 1
send_post_time = time.time() + sleep
while True:
# get video frame
img_idx = img_idx + 1
ret, input_img = cap.read()
img_h, img_w, _ = np.shape(input_img)
if img_idx == 1 or img_idx % skip_frame == 0:
time_detection = 0
time_network = 0
time_plot = 0
# detect faces using LBP detector
start_time = timeit.default_timer()
gray_img = cv2.cvtColor(input_img, cv2.COLOR_BGR2GRAY)
if mtccn:
detected = detector.detect_faces(input_img)
else:
detected = detector.detectMultiScale(gray_img, 1.1)
elapsed_time = timeit.default_timer() - start_time
time_detection = time_detection + elapsed_time
faces = np.empty((len(detected), img_size, img_size, 3))
input_img, time_network, time_plot, resultados = show_results(
detected, input_img, faces, ad, img_size, img_w, img_h, model,
model_gender, time_detection, time_network, time_plot, mtccn)
# Show the time cost (fps)
# print('time_detection:', time_detection)
# print('time_network:', time_network)
# print('time_plot:', time_plot)
# print('===============================')
cv2.waitKey(1)
if send_post_time < time.time():
send_post_time = time.time() + sleep
await nested(resultados)
def main():
weight_file = "../pre-trained/morph2/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
weight_file_gender = "../pre-trained/wiki_gender_models/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
face_cascade = cv2.CascadeClassifier('lbpcascade_frontalface_improved.xml')
try:
os.mkdir('./img')
except OSError:
pass
# load model and weights
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model_gender = SSR_net_general(img_size, stage_num, lambda_local,
lambda_d)()
model_gender.load_weights(weight_file_gender)
# capture video
cap = cv2.VideoCapture(0)
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1024 * 1)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 768 * 1)
img_idx = 0
detected = '' #make this not local variable
time_detection = 0
time_network = 0
time_plot = 0
skip_frame = 5 # every 5 frame do 1 detection and network forward propagation
ad = 0.5
while True:
# get video frame
ret, input_img = cap.read()
img_idx = img_idx + 1
img_h, img_w, _ = np.shape(input_img)
if img_idx == 1 or img_idx % skip_frame == 0:
time_detection = 0
time_network = 0
time_plot = 0
# detect faces using LBP detector
gray_img = cv2.cvtColor(input_img, cv2.COLOR_BGR2GRAY)
start_time = timeit.default_timer()
detected = face_cascade.detectMultiScale(gray_img, 1.1)
elapsed_time = timeit.default_timer() - start_time
time_detection = time_detection + elapsed_time
faces = np.empty((len(detected), img_size, img_size, 3))
input_img, time_network, time_plot = draw_results(
detected, input_img, faces, ad, img_size, img_w, img_h, model,
model_gender, time_detection, time_network, time_plot)
cv2.imwrite('img/' + str(img_idx) + '.png', input_img)
else:
input_img, time_network, time_plot = draw_results(
detected, input_img, faces, ad, img_size, img_w, img_h, model,
model_gender, time_detection, time_network, time_plot)
#Show the time cost (fps)
print('avefps_time_detection:', 1 / time_detection)
print('avefps_time_network:', skip_frame / time_network)
print('avefps_time_plot:', skip_frame / time_plot)
print('===============================')
key = cv2.waitKey(1)
def main():
#dynamicaly allocate GPU memory
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
K.tensorflow_backend.set_session(sess)
print('[LOAD DATA]')
images, ageLabel, genderLabel = prepData(64)
n_fold = 1
img_size = 64
stage_num = [3,3,3]
lambda_local = 1
lambda_d = 1
imdb_model = SSR_net(img_size,stage_num, lambda_local, lambda_d)()
imdb_model.compile(optimizer='adam', loss="mae", metrics=["mae"])
imdb_model.load_weights("imdb_age_ssrnet_3_3_3_64_1.0_1.0.h5")
imdb_model_gender = SSR_net_general(img_size,stage_num, lambda_local, lambda_d)()
imdb_model_gender.compile(optimizer='adam', loss="mae", metrics=["binary_accuracy"])
imdb_model_gender.load_weights("imdb_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
wiki_model = SSR_net(img_size,stage_num, lambda_local, lambda_d)()
wiki_model.compile(optimizer='adam', loss="mae", metrics=["mae"])
wiki_model.load_weights("wiki_age_ssrnet_3_3_3_64_1.0_1.0.h5")
wiki_model_gender = SSR_net_general(img_size,stage_num, lambda_local, lambda_d)()
wiki_model_gender.compile(optimizer='adam', loss="mae", metrics=["binary_accuracy"])
wiki_model_gender.load_weights("wiki_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
morph_model = SSR_net(img_size,stage_num, lambda_local, lambda_d)()
morph_model.compile(optimizer='adam', loss="mae", metrics=["mae"])
morph_model.load_weights("morph_age_ssrnet_3_3_3_64_1.0_1.0.h5")
morph_model_gender = SSR_net_general(img_size,stage_num, lambda_local, lambda_d)()
morph_model_gender.compile(optimizer='adam', loss="mae", metrics=["binary_accuracy"])
morph_model_gender.load_weights("morph_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
print('[K-FOLD] Started...')
kf = KFold(n_splits=10, shuffle=True, random_state=1)
kf_split = kf.split(ageLabel)
for _, test_idx in kf_split:
print('[K-FOLD] Fold {}'.format(n_fold))
testImages = images[test_idx]
testAge = ageLabel[test_idx]
testGender = genderLabel[test_idx]
scores = evaluate(imdb_model, testImages, testAge)
print('imdb Age score:', scores)
scores = evaluate(wiki_model, testImages, testAge)
print('wiki Age score:', scores)
scores = evaluate(morph_model, testImages, testAge)
print('morph Age score:', scores)
scores = evaluate(imdb_model_gender, testImages, testGender)
print('imdb Gender score:', scores)
scores = evaluate(wiki_model_gender, testImages, testGender)
print('wiki Gender score:', scores)
scores = evaluate(morph_model_gender, testImages, testGender)
print('morph Gender score:', scores)
n_fold += 1
del testImages, testAge, testGender, scores
def main():
args = get_args()
input_path1 = args.input1
input_path2 = args.input2
db_name = args.db
batch_size = args.batch_size
nb_epochs = args.nb_epochs
netType1 = args.netType1
netType2 = args.netType2
logging.debug("Loading training data...")
image1, age1, image_size = load_data_npz(input_path1)
logging.debug("Loading testing data...")
image2, age2, image_size = load_data_npz(input_path2)
start_decay_epoch = [30, 60]
optMethod = Adam()
stage_num = [3, 3, 3]
lambda_local = 0.25 * (netType1 % 5)
lambda_d = 0.25 * (netType2 % 5)
model = SSR_net(image_size, stage_num, lambda_local, lambda_d)()
save_name = 'ssrnet_%d_%d_%d_%d_%s_%s' % (stage_num[0], stage_num[1],
stage_num[2], image_size,
lambda_local, lambda_d)
model.compile(optimizer=optMethod, loss=["mae"], metrics={'pred_a': 'mae'})
if db_name == "megaage":
weight_file = "./pre-trained/wiki/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
#weight_file = "./pre-trained/imdb/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
model.load_weights(weight_file)
logging.debug("Model summary...")
model.count_params()
model.summary()
logging.debug("Saving model...")
mk_dir(db_name + "_models")
mk_dir(db_name + "_models/batch_size_%d/" % (batch_size))
mk_dir(db_name + "_models/batch_size_%d/" % (batch_size) + save_name)
mk_dir(db_name + "_checkpoints")
mk_dir(db_name + "_checkpoints/batch_size_%d/" % (batch_size))
plot_model(model,
to_file=db_name + "_models/batch_size_%d/" % (batch_size) +
save_name + "/" + save_name + ".png")
with open(
os.path.join(
db_name + "_models/batch_size_%d/" % (batch_size) + save_name,
save_name + '.json'), "w") as f:
f.write(model.to_json())
decaylearningrate = TYY_callbacks.DecayLearningRate(start_decay_epoch)
callbacks = [
ModelCheckpoint(db_name + "_checkpoints/batch_size_%d/" %
(batch_size) +
"weights.{epoch:02d}-{val_loss:.2f}.hdf5",
monitor="val_loss",
verbose=1,
save_best_only=True,
mode="auto"), decaylearningrate
]
logging.debug("Running training...")
data_num = len(image1) + len(image2)
indexes1 = np.arange(len(image1))
indexes2 = np.arange(len(image2))
np.random.shuffle(indexes1)
np.random.shuffle(indexes2)
x_train = image1[indexes1]
x_test = image2[indexes2]
y_train_a = age1[indexes1]
y_test_a = age2[indexes2]
train_num = len(image1)
hist = model.fit_generator(generator=data_generator_reg(
X=x_train, Y=y_train_a, batch_size=batch_size),
steps_per_epoch=train_num // batch_size,
validation_data=(x_test, [y_test_a]),
epochs=nb_epochs,
verbose=1,
callbacks=callbacks)
logging.debug("Saving weights...")
model.save_weights(os.path.join(
db_name + "_models/batch_size_%d/" % (batch_size) + save_name,
save_name + '.h5'),
overwrite=True)
pd.DataFrame(hist.history).to_hdf(
os.path.join(
db_name + "_models/batch_size_%d/" % (batch_size) + save_name,
'history_' + save_name + '.h5'), "history")
def main():
K.set_learning_phase(0) # make sure its testing mode
weight_file = "../pre-trained/wiki/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
weight_file_gender = "../pre-trained/wiki_gender_models/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
# for face detection
# detector = dlib.get_frontal_face_detector()
detector = MTCNN()
try:
os.mkdir('./img')
except OSError:
pass
# load model and weights
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model_gender = SSR_net_general(img_size, stage_num, lambda_local,
lambda_d)()
model_gender.load_weights(weight_file_gender)
clip = VideoFileClip(sys.argv[1]) # can be gif or movie
#python version
pyFlag = ''
if len(sys.argv) < 3:
pyFlag = '2' #default to use moviepy to show, this can work on python2.7 and python3.5
elif len(sys.argv) == 3:
pyFlag = sys.argv[2] #python version
else:
print('Wrong input!')
sys.exit()
img_idx = 0
detected = '' #make this not local variable
time_detection = 0
time_network = 0
time_plot = 0
ad = 0.4
skip_frame = 1 # every 5 frame do 1 detection and network forward propagation
for img in clip.iter_frames():
img_idx = img_idx + 1
input_img = img #using python2.7 with moivepy to show th image without channel flip
if pyFlag == '3':
input_img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_h, img_w, _ = np.shape(input_img)
input_img = cv2.resize(input_img, (1024, int(1024 * img_h / img_w)))
img_h, img_w, _ = np.shape(input_img)
if img_idx == 1 or img_idx % skip_frame == 0:
# detect faces using dlib detector
start_time = timeit.default_timer()
detected = detector.detect_faces(input_img)
elapsed_time = timeit.default_timer() - start_time
time_detection = time_detection + elapsed_time
faces = np.empty((len(detected), img_size, img_size, 3))
input_img, time_network, time_plot = draw_results(
detected, input_img, faces, ad, img_size, img_w, img_h, model,
model_gender, time_detection, time_network, time_plot)
#Show the time cost (fps)
# print('avefps_time_detection:',img_idx/time_detection)
# print('avefps_time_network:',img_idx/time_network)
# print('avefps_time_plot:',img_idx/time_plot)
# print('===============================')
if pyFlag == '3':
key = cv2.waitKey()
if key == 27:
break
age_net = None
# Load age and gender models
if (age_gender_kind == 'ssrnet'):
# Setup global parameters
face_size = 64
face_padding_ratio = 0.10
# Default parameters for SSR-Net
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
# Initialize gender net
gender_net = SSR_net_general(face_size, stage_num, lambda_local, lambda_d)()
gender_net.load_weights('age_gender_ssrnet_models/ssrnet_gender_3_3_3_64_1.0_1.0.h5')
# Initialize age net
age_net = SSR_net(face_size, stage_num, lambda_local, lambda_d)()
age_net.load_weights('age_gender_ssrnet_models/ssrnet_age_3_3_3_64_1.0_1.0.h5')
else:
# Setup global parameters
face_size = 227
face_padding_ratio = 0.0
# Initialize gender detector
gender_net = cv.dnn.readNetFromCaffe('gender_deploy.prototxt', 'gender_net.caffemodel')
# Initialize age detector
age_net = cv.dnn.readNetFromCaffe('age_deploy.prototxt', 'age_net.caffemodel')
# Mean values for gender_net and age_net
Genders = ['Male', 'Female']
Ages = ['(0-2)', '(4-6)', '(8-12)', '(15-20)', '(25-32)', '(38-43)', '(48-53)', '(60-100)']
def calculateParameters(height_orig, width_orig):
sess = K.get_session()
from tensorflow.python.framework import graph_util, graph_io
init_graph = sess.graph.as_graph_def()
main_graph = graph_util.convert_variables_to_constants(sess, init_graph, out_nodes)
graph_io.write_graph(main_graph, output_dir, name=model_name, as_text=False)
if log_tensorboard:
from tensorflow.python.tools import import_pb_to_tensorboard
import_pb_to_tensorboard.import_to_tensorboard(os.path.join(output_dir, model_name), output_dir)
# weight_file = "../pre-trained/morph2/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
weight_file = "../pre-trained/wiki/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
# weight_file = "../pre-trained/imdb/ssrnet_3_3_3_64_1.0_1.0/ssrnet_3_3_3_64_1.0_1.0.h5"
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
output_dir = os.path.join(os.getcwd(), "out")
keras_to_tensorflow(model, output_dir=output_dir, model_name="ssr_model.pb")
print("MODEL SAVED")
def check_inference_time():
age_layer = 'age_prediction'
gender_layer = 'gender_prediction'
logger.info('Load InceptionV3 model')
inceptionv3 = AgenderNetInceptionV3()
inceptionv3.setWeight('trainweight/inceptionv3_2/model.16-3.7887-0.9004-6.6744.h5')
inceptionv3_age = Model(inputs=inceptionv3.input,
outputs=inceptionv3.get_layer(age_layer).output)
inceptionv3_gender = Model(inputs=inceptionv3.input,
outputs=inceptionv3.get_layer(gender_layer).output)
logger.info('Load MobileNetV2 model')
mobilenetv2 = AgenderNetMobileNetV2()
mobilenetv2.setWeight('trainweight/mobilenetv2/model.10-3.8290-0.8965-6.9498.h5')
mobilenetv2_age = Model(inputs=mobilenetv2.input,
outputs=mobilenetv2.get_layer(age_layer).output)
mobilenetv2_gender = Model(inputs=mobilenetv2.input,
outputs=mobilenetv2.get_layer(gender_layer).output)
logger.info('Load SSRNet model')
ssrnet = SSRNet(64, [3, 3, 3], 1.0, 1.0)
ssrnet.setWeight('trainweight/agender_ssrnet/model.31-7.5452-0.8600-7.4051.h5')
ssrnet_age = Model(inputs=ssrnet.input,
outputs=ssrnet.get_layer(age_layer).output)
ssrnet_gender = Model(inputs=ssrnet.input,
outputs=ssrnet.get_layer(gender_layer).output)
logger.info('Load pretrain imdb model')
imdb_model = SSR_net(64, [3, 3, 3], 1.0, 1.0)()
imdb_model.load_weights("tes_ssrnet/imdb_age_ssrnet_3_3_3_64_1.0_1.0.h5")
imdb_model_gender = SSR_net_general(64, [3, 3, 3], 1.0, 1.0)()
imdb_model_gender.load_weights("tes_ssrnet/imdb_gender_ssrnet_3_3_3_64_1.0_1.0.h5")
images = cv2.imread('UTKface_aligned/part1/34_1_0_20170103183147490.jpg')
image = cv2.resize(images, (64, 64), interpolation = cv2.INTER_CUBIC)
X = image.astype('float16')
X = np.expand_dims(X, axis=0)
logger.info('Predict age and gender with SSR-Net')
wrapped = wrapper(predictone, ssrnet, X)
logger.info(proces_time(wrapped))
logger.info('Predict age with SSR-Net')
wrapped = wrapper(predictone, ssrnet_age, X)
logger.info(proces_time(wrapped))
logger.info('Predict gender with SSR-Net')
wrapped = wrapper(predictone, ssrnet_gender, X)
logger.info(proces_time(wrapped))
logger.info('Predict age with IMDB_SSR-Net')
wrapped = wrapper(predictone, imdb_model, X)
logger.info(proces_time(wrapped))
logger.info('Predict gender with IMDB_SSR-Net')
wrapped = wrapper(predictone, imdb_model_gender, X)
logger.info(proces_time(wrapped))
del X
image = cv2.resize(images, (96, 96), interpolation = cv2.INTER_CUBIC)
X = image.astype('float16')
X = np.expand_dims(X, axis=0)
logger.info('Predict age and gender with MobileNetV2')
wrapped = wrapper(predictone, mobilenetv2, X)
logger.info(proces_time(wrapped))
logger.info('Predict age with MobileNetV2')
wrapped = wrapper(predictone, mobilenetv2_age, X)
logger.info(proces_time(wrapped))
logger.info('Predict gender with MobileNetV2')
wrapped = wrapper(predictone, mobilenetv2_gender, X)
logger.info(proces_time(wrapped))
del X
X = images.astype('float16')
X = np.expand_dims(X, axis=0)
logger.info('Predict age and gender with InceptionV3')
wrapped = wrapper(predictone, inceptionv3, X)
logger.info(proces_time(wrapped))
logger.info('Predict age with InceptionV3')
wrapped = wrapper(predictone, inceptionv3_age, X)
logger.info(proces_time(wrapped))
logger.info('Predict gender with InceptionV3')
wrapped = wrapper(predictone, inceptionv3_gender, X)
logger.info(proces_time(wrapped))
lambda_local = 0.25*(netType1%5)
lambda_d = 0.25*(netType2%5)
logging.debug("Loading testing data...")
image2, age2, image_size = load_data_npz(test_file)
mk_dir('Results_csv')
model_file = 'megaage_models/batch_size_50/ssrnet_%d_%d_%d_%d_%s_%s/ssrnet_%d_%d_%d_%d_%s_%s.h5' % (stage_num[0],stage_num[1],stage_num[2], image_size, lambda_local, lambda_d, stage_num[0],stage_num[1],stage_num[2], image_size, lambda_local, lambda_d)
save_name = 'Results_csv/ssrnet_%d_%d_%d_%d_%s_%s_age.csv' % (stage_num[0],stage_num[1],stage_num[2], image_size, lambda_local, lambda_d)
'''''''''''''''''''''''''''''''''''''''''''''
load data
'''''''''''''''''''''''''''''''''''''''''''''
model = SSR_net(image_size,stage_num, lambda_local, lambda_d)()
logging.debug("Loading model file...")
model.load_weights(model_file)
age_p=model.predict(image2)
'''''''''''''''''''''''''''''''''''''''''''''
prediction
'''''''''''''''''''''''''''''''''''''''''''''
age_p2=age_p
pred=[['MAE'],[str(MAE(age2[age2>=-1],age_p2[age2>=-1]))],['CA3','CA5'],['0','0'],['ID','age','age_p','error']]
CA3=0
CA5=0
for i in range(0,len(image2)):
def main():
args = get_args()
input_path = args.input
db_name = args.db
batch_size = args.batch_size
nb_epochs = args.nb_epochs
validation_split = args.validation_split
netType1 = args.netType1
netType2 = args.netType2
logging.debug("Loading data...")
image, gender, age, image_size = load_data_npz(input_path)
x_data = image
y_data_a = age
start_decay_epoch = [30,60]
optMethod = Adam()
stage_num = [3,3,3]
lambda_local = 0.25*(netType1%5)
lambda_d = 0.25*(netType2%5)
model = SSR_net(image_size,stage_num, lambda_local, lambda_d)()
save_name = 'ssrnet_%d_%d_%d_%d_%s_%s' % (stage_num[0],stage_num[1],stage_num[2], image_size, lambda_local, lambda_d)
model.compile(optimizer=optMethod, loss=["mae"], metrics={'pred_a':'mae'})
if db_name == "wiki":
weight_file = f'imdb_models/{save_name}/{save_name}.h5'
if os.path.isfile(weight_file): # error out if file not exist
print("previous weight loading...")
model.load_weights(weight_file)
elif db_name == "morph":
weight_file = f'wiki_models/{save_name}/{save_name}.h5'
if os.path.isfile(weight_file):
print("previous weight loading...")
model.load_weights(weight_file)
logging.debug("Model summary...")
model.count_params()
model.summary()
logging.debug("Saving model...")
mk_dir(db_name+"_models")
mk_dir(db_name+"_models/"+save_name)
mk_dir(db_name+"_checkpoints")
plot_model(model, to_file=db_name+"_age_models/"+save_name+"/"+save_name+".png")
with open(os.path.join(db_name+"_age_models/"+save_name, save_name+'.json'), "w") as f:
f.write(model.to_json())
decaylearningrate = TYY_callbacks.DecayLearningRate(start_decay_epoch)
callbacks = [ModelCheckpoint(db_name+"_checkpoints/weights.{epoch:02d}-{val_loss:.2f}.hdf5",
monitor="val_loss",
verbose=1,
save_best_only=True,
mode="auto"), decaylearningrate
]
logging.debug("Running training...")
data_num = len(x_data)
indexes = np.arange(data_num)
np.random.shuffle(indexes)
x_data = x_data[indexes]
y_data_a = y_data_a[indexes]
train_num = int(data_num * (1 - validation_split))
x_train = x_data[:train_num]
x_test = x_data[train_num:]
y_train_a = y_data_a[:train_num]
y_test_a = y_data_a[train_num:]
hist = model.fit_generator(generator=data_generator_reg(X=x_train, Y=y_train_a, batch_size=batch_size),
steps_per_epoch=train_num // batch_size,
validation_data=(x_test, [y_test_a]),
epochs=nb_epochs, verbose=1,
callbacks=callbacks)
logging.debug("Saving weights...")
model.save_weights(os.path.join(db_name+"_models/"+save_name, save_name+'.h5'), overwrite=True)
pd.DataFrame(hist.history).to_hdf(os.path.join(db_name+"_models/"+save_name, 'history_'+save_name+'.h5'), "history")
emotion_target_size = emotion_classifier.input_shape[1:3]
# starting lists for calculating modes
emotion_window = []
# starting video streaming
cv2.namedWindow('window_frame')
video_capture = cv2.VideoCapture(0)
# load model and weights
img_size = 64
stage_num = [3, 3, 3]
lambda_local = 1
lambda_d = 1
model = SSR_net(img_size, stage_num, lambda_local, lambda_d)()
model.load_weights(weight_file)
model_gender = SSR_net_general(img_size, stage_num, lambda_local, lambda_d)()
model_gender.load_weights(weight_file_gender)
# Select video or webcam feed
cap = None
if (USE_WEBCAM == True):
cap = cv2.VideoCapture(0) # Webcam source
else:
cap = cv2.VideoCapture('./Emotion/demo/TGOP.mp4') # Video file source
while cap.isOpened(): # True:
ret, bgr_image = cap.read()
