def get_model(cfg_path):
# init
IOU_TH = 0.5
SCORE_TH = 0.4
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=IOU_TH,
score_th=SCORE_TH)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
#print("[*] load ckpt from {}.".format(tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
return model
def initialize():
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
# Setup logger
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# Define network
model = RetinaFaceModel(cfg, training=False,
iou_th=FLAGS.iou_th, score_th=FLAGS.score_th)
# Load checkpoints
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print(
"[*] load ckpt from {}.".format(tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
return model, cfg
def main(_argv):
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg, training=False, iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = FLAGS.weights
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
model.summary()
for i in model.layers:
print(i.output)
model.save(FLAGS.output)
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load model from weights.h5
# model.load_weights('./model/mbv2_weights.h5', by_name=True, skip_mismatch=True)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not FLAGS.webcam:
file_path = '/Users/lichaochao/Downloads/images_UMU/'
for file_name in os.listdir(file_path + 'source_images/'):
image_path = file_path + 'source_images/' + file_name
if not os.path.exists(image_path):
print(f"cannot find image path from {image_path}")
continue
img_raw = cv2.imread(image_path)
img = np.float32(img_raw.copy())
# img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
img_height, img_width, _ch = img.shape
# run model
outputs = model(img[np.newaxis, ...])
preds = tf.concat([
outputs[0][0], outputs[1][0, :, 1][..., tf.newaxis],
outputs[2][0, :, 1][..., tf.newaxis]
], -1)
priors = prior_box_tf((img_height, img_width), cfg['min_sizes'],
cfg['steps'], cfg['clip'])
decode_preds = decode_tf(preds, priors, cfg['variances'])
selected_indices = tf.image.non_max_suppression(
boxes=decode_preds[:, :4],
scores=decode_preds[:, -1],
max_output_size=tf.shape(decode_preds)[0],
iou_threshold=FLAGS.iou_th,
score_threshold=FLAGS.score_th)
outputs = tf.gather(decode_preds, selected_indices).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
has_face = False
is_smile = False
for prior_index in range(len(outputs)):
ann = outputs[prior_index]
if ann[-1] >= 0.5:
has_face = True
x1, y1 = int(ann[0] * img_width), int(ann[1] * img_height)
x2, y2 = int(ann[2] * img_width), int(ann[3] * img_height)
text = "face: {:.2f}".format(ann[-1] * 100)
cv2.putText(img, text, (x1 + 5, y1 - 10),
cv2.FONT_HERSHEY_DUPLEX, 0.5, (255, 255, 255))
if ann[-2] >= 0.5:
is_smile = True
smile_text = "smile: {:.2f}".format(ann[-2] * 100)
cv2.putText(img, smile_text, (x1 + 5, y1 + 30),
cv2.FONT_HERSHEY_DUPLEX, 0.5,
(255, 255, 255))
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
else:
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2)
if is_smile:
dst_file_path = file_path + '/smile_face/' + file_name
elif has_face:
dst_file_path = file_path + '/face/' + file_name
else:
dst_file_path = file_path + '/no_face/' + file_name
cv2.imwrite(dst_file_path, img)
print(dst_file_path)
else:
cam = cv2.VideoCapture('./data/linda_umu.mp4')
# cam.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
# cam.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
resize = FLAGS.down_scale_factor
frame_height = cam.get(cv2.CAP_PROP_FRAME_HEIGHT) * resize
frame_width = cam.get(cv2.CAP_PROP_FRAME_WIDTH) * resize
max_steps = max(cfg['steps'])
img_pad_h = max_steps - frame_height % max_steps if frame_height % max_steps > 0 else 0
img_pad_w = max_steps - frame_width % max_steps if frame_width % max_steps > 0 else 0
priors = prior_box_tf(
(frame_height + img_pad_h, frame_width + img_pad_w),
cfg['min_sizes'], cfg['steps'], cfg['clip'])
frame_index = 0
outputs = []
start_time = time.time()
while cam.isOpened():
_, frame = cam.read()
if frame is None:
print('no cam')
break
if frame_index < 5:
frame_index += 1
# continue
else:
frame_index = 0
img = np.float32(frame.copy())
if resize < 1:
img = cv2.resize(img, (0, 0),
fx=resize,
fy=resize,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max_steps)
# run model
outputs = model(img[np.newaxis, ...])
preds = tf.concat([
outputs[0][0], outputs[1][0, :, 1][..., tf.newaxis],
outputs[2][0, :, 1][..., tf.newaxis]
], -1)
decode_preds = decode_tf(preds, priors, cfg['variances'])
selected_indices = tf.image.non_max_suppression(
boxes=decode_preds[:, :4],
scores=decode_preds[:, -1],
max_output_size=tf.shape(decode_preds)[0],
iou_threshold=FLAGS.iou_th,
score_threshold=FLAGS.score_th)
outputs = tf.gather(decode_preds, selected_indices).numpy()
# recover padding effect
outputs = recover_pad_output(outputs,
pad_params,
resize=resize)
# calculate fps
fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
start_time = time.time()
cv2.putText(frame, fps_str, (25, 50), cv2.FONT_HERSHEY_DUPLEX,
0.75, (0, 0, 255), 2)
# draw results
for prior_index in range(len(outputs)):
draw_bbox_landm(frame, outputs[prior_index], frame_height,
frame_width)
# calculate fps
# fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
# start_time = time.time()
# cv2.putText(frame, fps_str, (25, 25),
# cv2.FONT_HERSHEY_DUPLEX, 0.75, (0, 255, 0), 2)
# show frame
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
exit()
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not FLAGS.webcam:
if not os.path.exists(FLAGS.img_path):
print(f"cannot find image path from {FLAGS.img_path}")
exit()
print("[*] Processing on single image {}".format(FLAGS.img_path))
img_raw = cv2.imread(FLAGS.img_path)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw and save results
save_img_path = os.path.join('out_' + os.path.basename(FLAGS.img_path))
for prior_index in range(len(outputs)):
draw_bbox_landm(img_raw, outputs[prior_index], img_height_raw,
img_width_raw)
cv2.imwrite(save_img_path, img_raw)
print(f"[*] save result at {save_img_path}")
else:
cam = cv2.VideoCapture("./videos/Bentall_Centra.MP4")
start_time = time.time()
while True:
_, frame = cam.read()
if frame is None:
print("no cam input")
frame_height, frame_width, _ = frame.shape
img = np.float32(frame.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw results
for prior_index in range(len(outputs)):
draw_bbox_landm(frame, outputs[prior_index], frame_height,
frame_width)
# calculate fps
fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
start_time = time.time()
cv2.putText(frame, fps_str, (25, 25), cv2.FONT_HERSHEY_DUPLEX,
0.75, (0, 255, 0), 2)
# show frame
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
exit()
from modules.models import RetinaFaceModel
from modules.utils import (set_memory_growth, load_yaml, draw_bbox_landm,
pad_input_image, recover_pad_output)
set_memory_growth()
# tf.debugging.set_log_device_placement(True)
cfg_path = './configs/retinaface_mbv2.yaml'
gpu = '0'
iou_th = 0.4
score_th = 0.5
cfg = load_yaml(cfg_path)
model = RetinaFaceModel(cfg, training=False, iou_th=iou_th, score_th=score_th)
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
def get_face_value(img_raw, down_scale_factor=0.3):
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load model from weights.h5
# model.load_weights('./model/mbv2_weights.h5', by_name=True, skip_mismatch=True)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not FLAGS.webcam:
if not os.path.exists(FLAGS.img_path):
print(f"cannot find image path from {FLAGS.img_path}")
exit()
print("[*] Processing on single image {}".format(FLAGS.img_path))
img_raw = cv2.imread(FLAGS.img_path)
img = np.float32(img_raw.copy())
# testing scale
target_size = 320
img_size_max = np.max(img.shape[0:2])
resize = float(target_size) / float(img_size_max)
img = cv2.resize(img,
None,
None,
fx=resize,
fy=resize,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
# draw and save results
save_img_path = os.path.join('out_' + os.path.basename(FLAGS.img_path))
for prior_index in range(len(outputs)):
draw_bbox_landm(img, outputs[prior_index], target_size,
target_size)
cv2.imwrite(save_img_path, img)
print(f"[*] save result at {save_img_path}")
else:
cam = cv2.VideoCapture('./data/lichaochao.mp4')
# cam = cv2.VideoCapture(0)
frame_height = int(cam.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_width = int(cam.get(cv2.CAP_PROP_FRAME_WIDTH))
fourcc = cv2.VideoWriter_fourcc(*'XVID')
fps = cam.get(cv2.CAP_PROP_FPS)
out = cv2.VideoWriter('chaochao1.mp4',
fourcc,
fps=fps,
frameSize=(frame_height, frame_width))
resize = FLAGS.down_scale_factor
frame_height *= resize
frame_width *= resize
max_steps = max(cfg['steps'])
img_pad_h = max_steps - frame_height % max_steps if frame_height % max_steps > 0 else 0
img_pad_w = max_steps - frame_width % max_steps if frame_width % max_steps > 0 else 0
priors = prior_box_tf(
(frame_height + img_pad_h, frame_width + img_pad_w),
cfg['min_sizes'], cfg['steps'], cfg['clip'])
frame_index = 0
outputs = []
start_time = time.time()
while cam.isOpened():
_, frame = cam.read()
if frame is None:
print('no cam')
break
if frame_index < 5:
frame_index += 1
# continue
else:
frame_index = 0
img = np.float32(frame.copy())
if resize < 1:
img = cv2.resize(img, (0, 0),
fx=resize,
fy=resize,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max_steps)
# run model
outputs = model(img[np.newaxis, ...])
preds = tf.concat([
outputs[0][0], outputs[1][0, :, 1][..., tf.newaxis],
outputs[2][0, :, 1][..., tf.newaxis]
], -1)
decode_preds = decode_tf(preds, priors, cfg['variances'])
selected_indices = tf.image.non_max_suppression(
boxes=decode_preds[:, :4],
scores=decode_preds[:, -1],
max_output_size=tf.shape(decode_preds)[0],
iou_threshold=FLAGS.iou_th,
score_threshold=FLAGS.score_th)
outputs = tf.gather(decode_preds, selected_indices).numpy()
# recover padding effect
outputs = recover_pad_output(outputs,
pad_params,
resize=resize)
# calculate fps
# fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
# start_time = time.time()
# cv2.putText(frame, fps_str, (25, 50),
# cv2.FONT_HERSHEY_DUPLEX, 0.75, (0, 0, 255), 2)
# draw results
for prior_index in range(len(outputs)):
draw_bbox_landm(frame, outputs[prior_index], frame_height,
frame_width)
# calculate fps
# fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
# start_time = time.time()
# cv2.putText(frame, fps_str, (25, 25),
# cv2.FONT_HERSHEY_DUPLEX, 0.75, (0, 255, 0), 2)
# show frame
out.write(frame)
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
exit()
from modules.models import RetinaFaceModel
from modules.utils import (set_memory_growth, load_yaml, draw_bbox_landm,
pad_input_image, recover_pad_output)
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg_path = './retinaface_mbv2.yaml'
cfg = load_yaml(cfg_path)
model = RetinaFaceModel(cfg, training=False, iou_th=0.4, score_th=0.5)
checkpoint_dir = './checkpoints/retinaface_mbv2'
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
app = Flask(__name__)
run_with_ngrok(app)
def main(_argv):
# init
face_aligner = FaceAligner()
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not FLAGS.webcam:
save_count = 0
for path, subdirs, files in os.walk(FLAGS.img_path):
for name in files:
if name.endswith('.jpg'):
img_path = os.path.join(path, name)
if not os.path.exists(img_path):
print(f"cannot find image path from {img_path}")
exit()
if save_count < FLAGS.img_num:
print("[*] Processing on single image {}".format(
img_path))
img_raw = cv2.imread(img_path)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img,
max_steps=max(
cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw and save results
save_img_path = os.path.join(
'out_' + os.path.basename(img_path))
for prior_index in range(len(outputs)):
draw_bbox_landm(img_raw, outputs[prior_index],
img_height_raw, img_width_raw)
cv2.imwrite(save_img_path, img_raw)
print(f"[*] save result at {save_img_path}")
save_count += 1
else:
cam = cv2.VideoCapture(0)
start_time = time.time()
while True:
_, frame = cam.read()
if frame is None:
print("no cam input")
frame_height, frame_width, _ = frame.shape
orig_frame = frame.copy()
face = None
img = cv2.resize(frame, (512, 512))
img = np.float32(frame.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
start_time = time.time()
outputs = model(img[np.newaxis, ...]).numpy()
inference_time = f"Inf: {time.time() - start_time}"
cv2.putText(frame, inference_time, (25, 50),
cv2.FONT_HERSHEY_DUPLEX, 0.75, (0, 255, 0), 2)
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw results
for prior_index in range(len(outputs)):
preds = decode_predictions((frame_width, frame_height),
outputs)
for key, value in preds.items():
bbox = value[0]['bbox']
left_eye = value[0]['left_eye']
right_eye = value[0]['right_eye']
# Our face ROI
face = orig_frame[bbox[1]:bbox[3], bbox[0]:bbox[2]]
# Eyes
x1_le = left_eye[0] - 25
y1_le = left_eye[1] - 25
x2_le = left_eye[0] + 25
y2_le = left_eye[1] + 25
x1_re = right_eye[0] - 25
y1_re = right_eye[1] - 25
x2_re = right_eye[0] + 25
y2_re = right_eye[1] + 25
if left_eye[1] > right_eye[1]:
A = (right_eye[0], left_eye[1])
else:
A = (left_eye[0], right_eye[1])
# Calc our rotating degree
delta_x = right_eye[0] - left_eye[0]
delta_y = right_eye[1] - left_eye[1]
angle = np.arctan(
delta_y / (delta_x + 1e-17)) # avoid devision by zero
angle = (angle * 180) / np.pi
# compute the desired right eye x-coordinate based on the
# desired x-coordinate of the left eye
desiredRightEyeX = 1.0 - 0.35
# determine the scale of the new resulting image by taking
# the ratio of the distance between eyes in the *current*
# image to the ratio of distance between eyes in the
# *desired* image
dist = np.sqrt((delta_x**2) + (delta_y**2))
desiredDist = (desiredRightEyeX - 0.35)
desiredDist *= 256
scale = desiredDist / dist
eyesCenter = ((left_eye[0] + right_eye[0]) // 2,
(left_eye[1] + right_eye[1]) // 2)
cv2.circle(frame, A, 5, (255, 0, 0), -1)
cv2.putText(frame, str(int(angle)), (x1_le - 15, y1_le),
cv2.FONT_HERSHEY_DUPLEX, 0.75, (0, 255, 0), 2)
cv2.line(frame, right_eye, left_eye, (0, 200, 200), 3)
cv2.line(frame, left_eye, A, (0, 200, 200), 3)
cv2.line(frame, right_eye, A, (0, 200, 200), 3)
cv2.line(frame, (left_eye[0], left_eye[1]),
(right_eye[0], right_eye[1]), (0, 200, 200), 3)
rotated = face_aligner.align(orig_frame, left_eye,
right_eye)
draw_bbox_landm(frame, outputs[prior_index], frame_height,
frame_width)
# calculate fps
fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
start_time = time.time()
cv2.putText(frame, fps_str, (25, 25), cv2.FONT_HERSHEY_DUPLEX,
0.75, (0, 255, 0), 2)
# show frame
cv2.imshow('frame', frame)
if face is not None:
cv2.imshow('face aligned', rotated)
if cv2.waitKey(1) == ord('q'):
exit()
def train_retinaface(cfg):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
if cfg['distributed']:
import horovod.tensorflow as hvd
# Initialize Horovod
hvd.init()
else:
hvd = []
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
reset_random_seeds()
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth(hvd)
# define network
model = RetinaFaceModel(cfg, training=True)
model.summary(line_length=80)
# define prior box
priors = prior_box((cfg['input_size'], cfg['input_size']),
cfg['min_sizes'], cfg['steps'], cfg['clip'])
# load dataset
train_dataset = load_dataset(cfg, priors, 'train', hvd)
if cfg['evaluation_during_training']:
val_dataset = load_dataset(cfg, priors, 'val', [])
# define optimizer
if cfg['distributed']:
init_lr = cfg['init_lr'] * hvd.size()
min_lr = cfg['min_lr'] * hvd.size()
steps_per_epoch = cfg['dataset_len'] // (cfg['batch_size'] * hvd.size())
else:
init_lr = cfg['init_lr']
min_lr = cfg['min_lr']
steps_per_epoch = cfg['dataset_len'] // cfg['batch_size']
learning_rate = MultiStepWarmUpLR(
initial_learning_rate=init_lr,
lr_steps=[e * steps_per_epoch for e in cfg['lr_decay_epoch']],
lr_rate=cfg['lr_rate'],
warmup_steps=cfg['warmup_epoch'] * steps_per_epoch,
min_lr=min_lr)
optimizer = tf.keras.optimizers.SGD(
learning_rate=learning_rate, momentum=0.9, nesterov=True)
# define losses function
multi_box_loss = MultiBoxLoss(num_class=cfg['num_class'])
# load checkpoint
checkpoint_dir = os.path.join(cfg['output_path'], 'checkpoints', cfg['sub_name'])
checkpoint = tf.train.Checkpoint(epoch=tf.Variable(0, name='epoch'),
optimizer=optimizer,
model=model)
manager = tf.train.CheckpointManager(checkpoint=checkpoint,
directory=checkpoint_dir,
max_to_keep=3)
os.makedirs(checkpoint_dir, exist_ok=True)
with open(os.path.join(checkpoint_dir, 'cfg.pickle'), 'wb') as handle:
pickle.dump(cfg, handle, protocol=pickle.HIGHEST_PROTOCOL)
if manager.latest_checkpoint:
checkpoint.restore(manager.latest_checkpoint)
print('[*] load ckpt from {}'.format(manager.latest_checkpoint))
else:
print("[*] training from scratch.")
# define training step function
@tf.function
def train_step(inputs, labels, first_batch, epoch):
with tf.GradientTape() as tape:
predictions = model(inputs, training=True)
losses = {}
losses['reg'] = tf.reduce_sum(model.losses)
losses['loc'], losses['landm'], losses['class'] = \
multi_box_loss(labels, predictions)
total_loss = tf.add_n([l for l in losses.values()])
if cfg['distributed']:
# Horovod: add Horovod Distributed GradientTape.
tape = hvd.DistributedGradientTape(tape)
grads = tape.gradient(total_loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
if cfg['distributed'] and first_batch and epoch:
hvd.broadcast_variables(model.variables, root_rank=0)
hvd.broadcast_variables(optimizer.variables(), root_rank=0)
return total_loss, losses
def test_step(inputs, img_name):
_, img_height_raw, img_width_raw, _ = inputs.shape
# pad input image to avoid unmatched shape problem
img = inputs[0].numpy()
# if img_name == '6_Funeral_Funeral_6_618':
# resize = 0.5 # this image is too big to avoid OOM problem
# img = cv2.resize(img, None, None, fx=resize, fy=resize,
# interpolation=cv2.INTER_LINEAR)
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
input_img = img[np.newaxis, ...]
predictions = model(input_img, training=False)
outputs = pred_to_outputs(cfg, predictions, input_img.shape).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
bboxs = outputs[:, :4]
confs = outputs[:, -1]
pred_boxes = []
for box, conf in zip(bboxs, confs):
x = int(box[0] * img_width_raw)
y = int(box[1] * img_height_raw)
w = int(box[2] * img_width_raw) - int(box[0] * img_width_raw)
h = int(box[3] * img_height_raw) - int(box[1] * img_height_raw)
pred_boxes.append([x, y, w, h, conf])
pred_boxes = np.array(pred_boxes).astype('float')
return pred_boxes
#training loop
summary_writer = tf.summary.create_file_writer(os.path.join(cfg['output_path'], 'logs', cfg['sub_name']))
prog_bar = ProgressBar(steps_per_epoch, 0)
if cfg['evaluation_during_training']:
widerface_eval_hard = WiderFaceEval(split='hard')
for epoch in range(cfg['epoch']):
try:
actual_epoch = epoch + 1
if cfg['distributed']:
if hvd.rank() == 0:
print("\nStart of epoch %d" % (actual_epoch,))
else:
print("\nStart of epoch %d" % (actual_epoch,))
checkpoint.epoch.assign_add(1)
start_time = time.time()
#Iterate over the batches of the dataset.
for batch, (x_batch_train, y_batch_train, img_name) in enumerate(train_dataset):
total_loss, losses = train_step(x_batch_train, y_batch_train, batch == 0, epoch == 0)
if cfg['distributed']:
if hvd.rank() == 0:
# prog_bar.update("epoch={}/{}, loss={:.4f}, lr={:.1e}".format(
# checkpoint.epoch.numpy(), cfg['epoch'], total_loss.numpy(), optimizer._decayed_lr(tf.float32)))
if batch % 100 == 0:
print("batch={}/{}, epoch={}/{}, loss={:.4f}, lr={:.1e}".format(
batch, steps_per_epoch, checkpoint.epoch.numpy(), cfg['epoch'], total_loss.numpy(), optimizer._decayed_lr(tf.float32)))
else:
prog_bar.update("epoch={}/{}, loss={:.4f}, lr={:.1e}".format(
checkpoint.epoch.numpy(), cfg['epoch'], total_loss.numpy(), optimizer._decayed_lr(tf.float32)))
# Display metrics at the end of each epoch.
# train_acc = train_acc_metric.result()
# print("\nTraining loss over epoch: %.4f" % (float(total_loss.numpy()),))
if cfg['distributed']:
if hvd.rank() == 0:
print("Time taken: %.2fs" % (time.time() - start_time))
manager.save()
print("\n[*] save ckpt file at {}".format(manager.latest_checkpoint))
else:
print("Time taken: %.2fs" % (time.time() - start_time))
manager.save()
print("\n[*] save ckpt file at {}".format(manager.latest_checkpoint))
if cfg['evaluation_during_training']:
# Run a validation loop at the end of each epoch.
for batch, (x_batch_val, y_batch_val, img_name) in enumerate(val_dataset.take(500)):
if '/' in img_name.numpy()[0].decode():
img_name = img_name.numpy()[0].decode().split('/')[1].split('.')[0]
else:
img_name = []
pred_boxes = test_step(x_batch_val, img_name)
gt_boxes = labels_to_boxes(y_batch_val)
widerface_eval_hard.update(pred_boxes, gt_boxes, img_name)
ap_hard = widerface_eval_hard.calculate_ap()
widerface_eval_hard.reset()
if cfg['distributed']:
if hvd.rank() == 0:
print("Validation acc: %.4f" % (float(ap_hard),))
else:
print("Validation acc: %.4f" % (float(ap_hard),))
def tensorboard_writer():
with summary_writer.as_default():
tf.summary.scalar('loss/total_loss', total_loss, step=actual_epoch)
for k, l in losses.items():
tf.summary.scalar('loss/{}'.format(k), l, step=actual_epoch)
tf.summary.scalar('learning_rate', optimizer._decayed_lr(tf.float32), step=actual_epoch)
if cfg['evaluation_during_training']:
tf.summary.scalar('Val AP', ap_hard, step=actual_epoch)
if cfg['distributed']:
if hvd.rank() == 0:
tensorboard_writer()
else:
tensorboard_writer()
except Exception as E:
print(E)
continue
if cfg['distributed']:
if hvd.rank() == 0:
manager.save()
print("\n[*] training done! save ckpt file at {}".format(
manager.latest_checkpoint))
else:
manager.save()
print("\n[*] training done! save ckpt file at {}".format(
manager.latest_checkpoint))
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
#print("[*] load ckpt from {}.".format(tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not os.path.exists(FLAGS.img_path):
print(f"cannot find image path from {FLAGS.img_path}")
exit()
print("[*] Processing on single image {}".format(FLAGS.img_path))
img_raw = cv2.imread(FLAGS.img_path)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw and save results
imgs = []
DIM = 64
save_img_path = os.path.join('data/out_' +
os.path.basename(FLAGS.img_path))
for prior_index in range(9):
if (prior_index < len(outputs)):
img = get_bbox_imgs(img_raw, outputs[prior_index], img_height_raw,
img_width_raw)
img = cv2.resize(img, (DIM, DIM))
imgs.append(img)
else:
imgs.append(Image.new('RGB', (DIM, DIM)))
imga = imgs[0]
for img in imgs[1:3]:
imga = np.concatenate((imga, img), axis=1)
imgb = imgs[3]
for img in imgs[4:6]:
imgb = np.concatenate((imgb, img), axis=1)
imgf = np.concatenate((imga, imgb), axis=0)
imgc = imgs[6]
for img in imgs[7:9]:
imgc = np.concatenate((imgc, img), axis=1)
imgf = np.concatenate((imgf, imgc), axis=0)
cv2.imwrite(save_img_path, imgf)
print(f"[*] save result at {save_img_path}")
def main(_argv):
# init
CONFIG_PATH = './configs/retinaface_mbv2.yaml'
GPU = '0'
IOU_TH = 0.4
SCORE_TH = 0.5
WEBCAM = False
DOWN_FACTOR = 1.0
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = GPU
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(CONFIG_PATH)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=IOU_TH,
score_th=SCORE_TH)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
return model
if not WEBCAM:
# if not os.path.exists(IMG_PATH):
# print(f"cannot find image path from {IMG_PATH}")
# exit()
# print("[*] Processing on single image {}".format(IMG_PATH))
# img_raw = cv2.imread(IMG_PATH)
img_raw = input_image
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
if DOWN_FACTOR < 1.0:
img = cv2.resize(img, (0, 0),
fx=DOWN_FACTOR,
fy=DOWN_FACTOR,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw and save results
save_img_path = "result.jpg"
for prior_index in range(len(outputs)):
draw_bbox_landm(img_raw, outputs[prior_index], img_height_raw,
img_width_raw)
cv2.imwrite(save_img_path, img_raw)
print(f"[*] save result at {save_img_path}")
else:
cam = cv2.VideoCapture(0)
start_time = time.time()
while True:
_, frame = cam.read()
if frame is None:
print("no cam input")
frame_height, frame_width, _ = frame.shape
img = np.float32(frame.copy())
if DOWN_FACTOR < 1.0:
img = cv2.resize(img, (0, 0),
fx=DOWN_FACTOR,
fy=DOWN_FACTOR,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw results
for prior_index in range(len(outputs)):
draw_bbox_landm(frame, outputs[prior_index], frame_height,
frame_width)
# calculate fps
fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
start_time = time.time()
cv2.putText(frame, fps_str, (25, 25), cv2.FONT_HERSHEY_DUPLEX,
0.75, (0, 255, 0), 2)
# show frame
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
exit()
return outputs
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
if not FLAGS.webcam:
if not os.path.exists(FLAGS.img_path):
print(f"cannot find image path from {FLAGS.img_path}")
exit()
print("[*] Processing on single image {}".format(FLAGS.img_path))
img_raw = cv2.imread(FLAGS.img_path)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw and save results
save_img_path = os.path.join('out_' + os.path.basename(FLAGS.img_path))
for prior_index in range(len(outputs)):
draw_bbox_landm(img_raw, outputs[prior_index], img_height_raw,
img_width_raw)
cv2.imwrite(save_img_path, img_raw)
print(f"[*] save result at {save_img_path}")
else:
cam = cv2.VideoCapture(FLAGS.vid_path)
fps = int(cam.get(cv2.CAP_PROP_FPS))
### Saving Video to file
frame_width = int(cam.get(3))
frame_height = int(cam.get(4))
# Define the codec and create VideoWriter object.The output is stored in 'outpy.avi' file.
# import os
if not os.path.exists('./output'):
print('Creating folder: output/ for saving video.')
os.makedirs('./output')
out = cv2.VideoWriter('./output/output.avi',
cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 10,
(frame_width, frame_height))
start_time = time.time()
counter = 0
frameCount = 0
while True:
_, frame = cam.read()
if frame is None:
print("no cam input")
frameCount = frameCount + 1
frame_height, frame_width, _ = frame.shape
img = np.float32(frame.copy())
if FLAGS.down_scale_factor < 1.0:
img = cv2.resize(img, (0, 0),
fx=FLAGS.down_scale_factor,
fy=FLAGS.down_scale_factor,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# draw results
for prior_index in range(len(outputs)):
croppedFace = draw_bbox_landm(frame, outputs[prior_index],
frame_height, frame_width)
if frameCount >= fps * FLAGS.dfps:
fileName = "%d.png" % counter
cv2.imwrite(FLAGS.dst_path + fileName, croppedFace)
print('Saved:', fileName)
counter = counter + 1
frameCount = 0
# calculate fps
fps_str = "FPS: %.2f" % (1 / (time.time() - start_time))
start_time = time.time()
cv2.putText(frame, fps_str, (25, 25), cv2.FONT_HERSHEY_DUPLEX,
0.75, (0, 255, 0), 2)
# show frame
if FLAGS.preview:
cv2.imshow('frame', frame)
out.write(frame)
if cv2.waitKey(1) == ord('q'):
exit()
def main(_argv):
# FUNCTIONS FOR CROPPING
#####################################################################################################
def bounding_box(img, ann, img_height, img_width):
x1, y1, x2, y2 = int(ann[0] * img_width), int(ann[1] * img_height), \
int(ann[2] * img_width), int(ann[3] * img_height)
return x1, y1, x2, y2
def calc_points(x, y, side):
return int(x - side / 2), int(x +
side / 2), int(y -
side / 2), int(y +
side / 2)
def adjust_points(x_center, y_center, original_longest, scaling_factor,
min_scaling_factor):
factors = np.arange(scaling_factor, min_scaling_factor - 0.04, -0.05)
for factor in factors:
# calculate nex points
x1, x2, y1, y2 = calc_points(x_center, y_center,
int(original_longest * factor))
for i in range(FLAGS.max_iter):
if x1 < 0:
x2 -= x1
x1 = 0
if y1 < 0:
y2 -= y1
y1 = 0
if x2 > img_raw.shape[1]:
x1 -= x2
x2 = img_raw.shape[1]
if y2 > img_raw.shape[0]:
y1 -= y2
y2 = img_raw.shape[0]
if x1 >= 0 and y1 >= 0 and x2 <= img_raw.shape[
1] and y2 <= img_raw.shape[0]:
return x1, x2, y1, y2, True
print("Not cropping", img_path,
"due to a problem with a cropping square box")
return x1, x2, y1, y2, False
def get_dim(lst):
return [(lst[3] - lst[1]) * (lst[2] - lst[0])]
def get_max(outputs, lst):
area = [i[0] for i in lst]
prob = [i[1] for i in lst]
max_area_index = set([i for i, j in enumerate(area) if j == max(area)])
max_prob_index = set([i for i, j in enumerate(prob) if j == max(prob)])
indecies = list(max_area_index.intersection(max_prob_index))
if len(indecies) >= 1: return [outputs[indecies[0]]]
elif len(indecies
) == 0: # if there is a mismatch, return the largest element
if len(list(max_area_index)) >= 1:
return [outputs[list(max_area_index)[0]]]
else: # precautionary because there should always be at least one face
print("Not cropping", img_path,
"due to a problem with returning the largest element")
return []
#####################################################################################################
# MODEL
#####################################################################################################
# initialisation
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
#####################################################################################################
# CROPPING
#####################################################################################################
# check if the path exits
if not os.path.exists(FLAGS.path):
print(f"cannot find the specified path from {FLAGS.path}")
exit()
# make a corresponding directory
try:
os.mkdir(FLAGS.path.replace("images", "cropped_images"))
except FileExistsError:
print(FLAGS.path.replace("images", "cropped_images"), "already exists")
# eget subdirectories within the specified folder
subdirectories = [FLAGS.path+'/'+i for i in os.listdir(FLAGS.path) \
if os.path.isdir(FLAGS.path+'/'+i)]
# loop through each folder
for subdir in sorted(subdirectories):
# create corresponding folders for cropped data and get all images in a given folder
if 'original' in subdir: x = 3
else: x = 7
try:
os.mkdir(subdir.replace("images", "cropped_images"))
images_lst = glob.glob(subdir + "/*.png")
cropped_images_lst = []
print(subdir[len(subdir) - x:len(subdir)])
except FileExistsError:
# count number of existing images in this subdirectory, if same as original, skip
images_lst = glob.glob(subdir + "/*.png")
cropped_images_lst = glob.glob(
subdir.replace("images", "cropped_images") + "/*.png")
cropped_images_lst = [
e[len(e) - 8:len(e)] for e in cropped_images_lst
]
if len(images_lst) == len(cropped_images_lst):
print(subdir[len(subdir) - x:len(subdir)],
"has already been generated")
continue
else:
print(subdir[len(subdir) - x:len(subdir)])
# loop through each image in a given folder
for img_path in sorted(images_lst):
if img_path[len(img_path) - 8:len(img_path)] in cropped_images_lst:
continue
img_raw = cv2.imread(img_path)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image (unmatched shape problem), run model, recover padding effect
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
outputs = model(img[np.newaxis, ...]).numpy()
outputs = recover_pad_output(outputs, pad_params)
# get rid of elements which are faces with less that threshold probability
outputs = [i for i in outputs if i[15] >= FLAGS.threshold_prob]
# flag any images which have no recognised faces in them
if len(outputs) == 0:
print("no faces detected for", img_path)
# if more than one face detected, select the largest and most definite
elif len(outputs) > 1:
f = [list(bounding_box(img_raw, i[0:4], img_height_raw, img_width_raw)) + [i[15]] \
for i in outputs]
f = [get_dim(i[0:4]) + [i[4]] for i in f]
outputs = get_max(outputs, f)
# keeping as a loop in case we decide to use multiple faces per frame in the future
# get cropping coordinates and save results
for prior_index in range(len(outputs)):
# get the bounding box coordinates
bb_x1, bb_y1, bb_x2, bb_y2 = bounding_box(
img_raw, outputs[prior_index], img_height_raw,
img_width_raw)
# scale up the magnitude of the longest side
original_longest = int(max(bb_x2 - bb_x1, bb_y2 - bb_y1))
longest = int(original_longest * FLAGS.scaling_factor)
x_center = int((bb_x1 + bb_x2) / 2)
y_center = int((bb_y1 + bb_y2) / 2)
x1, x2, y1, y2, save_image = adjust_points(
x_center, y_center, original_longest, FLAGS.scaling_factor,
FLAGS.min_scaling_factor)
if save_image:
try:
save_img_path = os.path.join(subdir.replace("images", "cropped_images") \
+ "/" + img_path.replace(subdir + '/', ''))
cv2.imwrite(save_img_path, img_raw[y1:y2, x1:x2])
except:
print(img_path, "is not cropped for unknown reasons")
def main(_):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg, training=True)
model.summary(line_length=80)
# define prior box
priors = prior_box((cfg['input_size'], cfg['input_size']),
cfg['min_sizes'], cfg['steps'], cfg['clip'])
# load dataset
train_dataset = load_dataset(cfg, priors, shuffle=True)
# define optimizer
steps_per_epoch = cfg['dataset_len'] // cfg['batch_size']
learning_rate = MultiStepWarmUpLR(
initial_learning_rate=cfg['init_lr'],
lr_steps=[e * steps_per_epoch for e in cfg['lr_decay_epoch']],
lr_rate=cfg['lr_rate'],
warmup_steps=cfg['warmup_epoch'] * steps_per_epoch,
min_lr=cfg['min_lr'])
optimizer = tf.keras.optimizers.SGD(learning_rate=learning_rate,
momentum=0.9,
nesterov=True)
# define losses function
multi_box_loss = MultiBoxLoss()
# load checkpoint
checkpoint_dir = '/content/drive/My Drive/Colab/checkpoints/' + cfg[
'sub_name']
checkpoint = tf.train.Checkpoint(step=tf.Variable(0, name='step'),
optimizer=optimizer,
model=model)
manager = tf.train.CheckpointManager(checkpoint=checkpoint,
directory=checkpoint_dir,
max_to_keep=3)
if manager.latest_checkpoint:
checkpoint.restore(manager.latest_checkpoint)
print('[*] load ckpt from {} at step {}.'.format(
manager.latest_checkpoint, checkpoint.step.numpy()))
else:
print("[*] training from scratch.")
# define training step function
@tf.function
def train_step(inputs, labels):
with tf.GradientTape() as tape:
predictions = model(inputs, training=True)
losses = {}
losses['reg'] = tf.reduce_sum(model.losses)
losses['loc'], losses['landm'], losses['class'] = \
multi_box_loss(labels, predictions)
total_loss = tf.add_n([l for l in losses.values()])
grads = tape.gradient(total_loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
return total_loss, losses
# training loop
summary_writer = tf.summary.create_file_writer('./logs/' + cfg['sub_name'])
remain_steps = max(
steps_per_epoch * cfg['epoch'] - checkpoint.step.numpy(), 0)
prog_bar = ProgressBar(steps_per_epoch,
checkpoint.step.numpy() % steps_per_epoch)
for inputs, labels in train_dataset.take(remain_steps):
checkpoint.step.assign_add(1)
steps = checkpoint.step.numpy()
total_loss, losses = train_step(inputs, labels)
prog_bar.update("epoch={}/{}, loss={:.4f}, lr={:.1e}".format(
((steps - 1) // steps_per_epoch) + 1, cfg['epoch'],
total_loss.numpy(),
optimizer.lr(steps).numpy()))
if steps % 10 == 0:
with summary_writer.as_default():
tf.summary.scalar('loss/total_loss', total_loss, step=steps)
for k, l in losses.items():
tf.summary.scalar('loss/{}'.format(k), l, step=steps)
tf.summary.scalar('learning_rate',
optimizer.lr(steps),
step=steps)
if steps % cfg['save_steps'] == 0:
manager.save()
print("\n[*] save ckpt file at {}".format(
manager.latest_checkpoint))
manager.save()
print("\n[*] training done! save ckpt file at {}".format(
manager.latest_checkpoint))
def main(_argv):
# init
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
os.environ["CUDA_VISIBLE_DEVICES"] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg,
training=False,
iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint
checkpoint_dir = "./checkpoints/" + cfg["sub_name"]
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
# evaluation on testing dataset
testset_folder = cfg["testing_dataset_path"]
testset_list = os.path.join(testset_folder, "label.txt")
img_paths, _ = load_info(testset_list)
for img_index, img_path in enumerate(img_paths):
print(" [{} / {}] det {}".format(img_index + 1, len(img_paths),
img_path))
img_raw = cv2.imread(img_path, cv2.IMREAD_COLOR)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
# testing scale
target_size = 1600
max_size = 2150
img_shape = img.shape
img_size_min = np.min(img_shape[0:2])
img_size_max = np.max(img_shape[0:2])
resize = float(target_size) / float(img_size_min)
# prevent bigger axis from being more than max_size:
if np.round(resize * img_size_max) > max_size:
resize = float(max_size) / float(img_size_max)
if FLAGS.origin_size:
if os.path.basename(img_path) == "6_Funeral_Funeral_6_618.jpg":
resize = 0.5 # this image is too big to avoid OOM problem
else:
resize = 1
img = cv2.resize(img,
None,
None,
fx=resize,
fy=resize,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image_(img, max_steps=max(cfg["steps"]))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# write results
img_name = os.path.basename(img_path)
sub_dir = os.path.basename(os.path.dirname(img_path))
save_name = os.path.join(FLAGS.save_folder, sub_dir,
img_name.replace(".jpg", ".txt"))
pathlib.Path(os.path.join(FLAGS.save_folder,
sub_dir)).mkdir(parents=True, exist_ok=True)
with open(save_name, "w") as file:
bboxs = outputs[:, :4]
confs = outputs[:, -1]
file_name = img_name + "\n"
bboxs_num = str(len(bboxs)) + "\n"
file.write(file_name)
file.write(bboxs_num)
for box, conf in zip(bboxs, confs):
x = int(box[0] * img_width_raw)
y = int(box[1] * img_height_raw)
w = int(box[2] * img_width_raw) - int(box[0] * img_width_raw)
h = int(box[3] * img_height_raw) - int(box[1] * img_height_raw)
confidence = str(conf)
line = str(x) + " " + str(y) + " " + str(w) + " " + str(
h) + " " + confidence + " \n"
file.write(line)
# save images
pathlib.Path(os.path.join("./results", cfg["sub_name"],
sub_dir)).mkdir(parents=True, exist_ok=True)
if FLAGS.save_image:
for prior_index in range(len(outputs)):
if outputs[prior_index][15] >= FLAGS.vis_th:
draw_bbox_landm(img_raw, outputs[prior_index],
img_height_raw, img_width_raw)
cv2.imwrite(
os.path.join("./results", cfg["sub_name"], sub_dir, img_name),
img_raw)
def main(_argv):
# init
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['CUDA_VISIBLE_DEVICES'] = FLAGS.gpu
logger = tf.get_logger()
logger.disabled = True
logger.setLevel(logging.FATAL)
set_memory_growth()
cfg = load_yaml(FLAGS.cfg_path)
# define network
model = RetinaFaceModel(cfg, training=False, iou_th=FLAGS.iou_th,
score_th=FLAGS.score_th)
# load checkpoint12
checkpoint_dir = './checkpoints/' + cfg['sub_name']
checkpoint = tf.train.Checkpoint(model=model)
if tf.train.latest_checkpoint(checkpoint_dir):
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
print("[*] load ckpt from {}.".format(
tf.train.latest_checkpoint(checkpoint_dir)))
else:
print("[*] Cannot find ckpt from {}.".format(checkpoint_dir))
exit()
# evaluation on testing dataset
testing_dataset_path = cfg['testing_dataset_path']
img_paths, _ = load_info(testing_dataset_path, './data/CelebA/train_labels.txt')
for img_index, img_path in enumerate(img_paths):
print(" [{} / {}] det {}".format(img_index + 1, len(img_paths),
img_path))
img_raw = cv2.imread(img_path, cv2.IMREAD_COLOR)
img_height_raw, img_width_raw, _ = img_raw.shape
img = np.float32(img_raw.copy())
# testing scale
if not FLAGS.origin_size:
target_size = 320
img_size_max = np.max(img.shape[0:2])
resize = float(target_size) / float(img_size_max)
img = cv2.resize(img, None, None, fx=resize, fy=resize,
interpolation=cv2.INTER_LINEAR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pad input image to avoid unmatched shape problem
img, pad_params = pad_input_image(img, max_steps=max(cfg['steps']))
# run model
outputs = model(img[np.newaxis, ...]).numpy()
# recover padding effect
outputs = recover_pad_output(outputs, pad_params)
# write results
img_name = os.path.basename(img_path)
sub_dir = os.path.basename(os.path.dirname(img_path))
save_name = os.path.join(
FLAGS.save_folder, sub_dir, img_name.replace('.jpg', '.txt'))
pathlib.Path(os.path.join(FLAGS.save_folder, sub_dir)).mkdir(
parents=True, exist_ok=True)
with open(save_name, "w") as file:
bboxs = outputs[:, :4]
smile_confs = outputs[:, 4]
face_confs = outputs[:, -1]
file_name = img_name + "\n"
bboxs_num = str(len(bboxs)) + "\n"
file.write(file_name)
file.write(bboxs_num)
for box, smile_conf, face_conf in zip(bboxs, smile_confs, face_confs):
x = int(box[0] * img_width_raw)
y = int(box[1] * img_height_raw)
w = int(box[2] * img_width_raw) - int(box[0] * img_width_raw)
h = int(box[3] * img_height_raw) - int(box[1] * img_height_raw)
line = str(x) + " " + str(y) + " " + str(w) + " " + str(h) + " " + str(smile_conf) + " " + str(
face_conf) + " \n"
file.write(line)
# save images
pathlib.Path(os.path.join(
'./results', cfg['sub_name'], sub_dir)).mkdir(
parents=True, exist_ok=True)
if FLAGS.save_image:
for prior_index in range(len(outputs)):
if outputs[prior_index][-1] >= FLAGS.vis_th:
draw_bbox_landm(img_raw, outputs[prior_index],
img_height_raw, img_width_raw)
cv2.imwrite(os.path.join('./results', cfg['sub_name'], sub_dir,
img_name), img_raw)
