def main(model_path, img_path, backend):
fashion_model = FashionClassifier(model_path, backend, 1.0, (32, 32), (), False)
avg_time = 0.0
if os.path.isdir(img_path):
images = os.listdir(img_path)
folder_size = len(images)
total_time = 0.0
for img in images:
image = cv2.imread(img_path+"/"+img)
t1 = cv2.TickMeter()
t1.start()
print(img, " ", fashion_model.predict(image))
t1.stop()
print(t1.getTimeMilli())
total_time = total_time + t1.getTimeMilli()
avg_time = total_time / folder_size
else:
t1 = cv2.TickMeter()
t1.start()
image = cv2.imread(img_path)
t1.stop()
avg_time = t1.getTimeMilli()
image_name = img_path.split("/")[len(img_path.split("/"))-1]
print(image_name, " ", fashion_model.predict(image))
print("\n AVG TIME PER IMAGE: %.2f" % avg_time + "ms")
def noise_estimate(im, pch_size=8):
'''
Implement of noise level estimation of the following paper:
Chen G , Zhu F , Heng P A . An Efficient Statistical Method for Image Noise Level Estimation[C]// 2015 IEEE International Conference
on Computer Vision (ICCV). IEEE Computer Society, 2015.
Input:
im: the noise image, H x W x 3 or H x W numpy tensor, range [0,1]
pch_size: patch_size
Output:
noise_level: the estimated noise level
'''
if im.ndim == 3:
im = im.transpose((2, 0, 1))
else:
im = np.expand_dims(im, axis=0)
# image to patch
pch = im2patch(im, pch_size,
16) # C x pch_size x pch_size x num_pch tensor
num_pch = pch.shape[3]
pch = pch.reshape((-1, num_pch)) # d x num_pch matrix
d = pch.shape[0]
# pch = np.ones([32400,64], np.float32)
#mu = pch.mean(axis=1, keepdims=True) # d x 1
tm1 = cv2.TickMeter()
tm1.start()
#mean,eigenvectors,eigenvalues = cv2.PCACompute2(pch.transpose(),mean=None)
tm1.stop()
tm = cv2.TickMeter()
#start =cv2.getTickCount()
#mu = pch.mean(axis=1, keepdims=True) # d x 1
#XX = np.ones([32400,64], np.float32)
tm.start()
mu = pch.mean(axis=1, keepdims=True) # d x 1
X = pch - mu
sigma_X = np.matmul(X, X.transpose()) / num_pch
sig_value, _ = np.linalg.eigh(sigma_X)
tm.stop()
tcv = tm.getTimeMilli()
tcv1 = tm1.getTimeMilli()
print('tcv {}ms, tcv1 {} ms', tcv / 100, tcv1 / 100)
sig_value.sort()
for ii in range(-1, -d - 1, -1):
tau = np.mean(sig_value[:ii])
if np.sum(sig_value[:ii] > tau) == np.sum(sig_value[:ii] < tau):
return np.sqrt(tau)
def __init__(self, source=0):
self._stream = cv.VideoCapture(source)
self._counter = cv.TickMeter()
self._frames = 0
self._runtime = 0
self._fps = 0
self._frame_time = 0
def DetectQRFrmImage(self, inputfile):
inputimg = cv.imread(inputfile, cv.IMREAD_COLOR)
if inputimg is None:
print('ERROR: Can not read image: {}'.format(inputfile))
return
print('Run {:s} on image [{:d}x{:d}]'.format(self.getQRModeString(),
inputimg.shape[1],
inputimg.shape[0]))
qrCode = cv.QRCodeDetector()
count = 10
timer = cv.TickMeter()
for _ in range(count):
timer.start()
points, decode_info = self.runQR(qrCode, inputimg)
timer.stop()
fps = count / timer.getTimeSec()
print('FPS: {}'.format(fps))
result = inputimg
self.drawQRCodeResults(result, points, decode_info, fps)
cv.imshow("QR", result)
cv.waitKey(1)
if self.out != '':
outfile = self.fname + self.fext
print("Saving Result: {}".format(outfile))
cv.imwrite(outfile, result)
print("Press any key to exit ...")
cv.waitKey(0)
print("Exit")
def init_model(transform):
global mark_detector, box_process, img_queue, box_queue, pose_estimator, pose_stabilizers, tm
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
# img_queue.put(sample_frame)
box_process = Process(target=get_face,
args=(
mark_detector,
img_queue,
box_queue,
))
box_process.start()
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
return None, None
def test_one(title, a, b):
# 创建类
if "AverageHash" == title:
hashFun = cv2.img_hash.AverageHash_create()
elif "PHash" == title:
hashFun = cv2.img_hash.PHash_create()
elif "MarrHildrethHash" == title:
hashFun = cv2.img_hash.MarrHildrethHash_create()
elif "RadialVarianceHash" == title:
hashFun = cv2.img_hash.RadialVarianceHash_create()
elif "BlockMeanHash" == title:
hashFun = cv2.img_hash.BlockMeanHash_create()
elif "ColorMomentHash" == title:
hashFun = cv2.img_hash.ColorMomentHash_create()
tick = cv2.TickMeter()
print("=== " + title + " ===")
tick.reset()
tick.start()
# # 计算图a的哈希值
hashA = hashFun.compute(a)
tick.stop()
print("compute1: " + str(tick.getTimeMilli()) + " ms")
tick.reset()
tick.start()
# 计算图b的哈希值
hashB = hashFun.compute(b)
tick.stop()
print("compute2: " + str(tick.getTimeMilli()) + " ms")
# 比较两张图像哈希值的距离
print("compare: " + str(hashFun.compare(hashA, hashB)))
def time_inverse():
src = cv2.imread('lenna.bmp', cv2.IMREAD_GRAYSCALE)
print(src.shape)
print(src)
if src is None :
print('Image load failed!')
return
dst = np.empty(src.shape, dtype = src.dtype)
tm = cv2.TickMeter()
tm.start()
for y in range(src.shape[0]):
for x in range(src.shape[1]):
dst[y, x] = 255 - src[y, x]
tm.stop()
print('Image inverse implementation took %4.3f ms' % tm.getTimeMilli())
print(dst)
cv2.imshow('src', src)
cv2.imshow('dst', dst)
cv2.waitKey()
cv2.destroyAllWindows()
def speed_test():
images = get_images()
detect_time = 0
detect_num = 0
timer = cv2.TickMeter()
predictor = Predictor()
# skip first predict because it may go longer
predictor.predict_bounding_boxes(images[0])
for image in images:
image = cv2.resize(image, (640, 480))
timer.start()
predictor.predict_bounding_boxes(image)
timer.stop()
detect_time += timer.getTimeMilli()
detect_num += 1
timer.reset()
average_time = detect_time / detect_num
print("average mtcnn prediction_time {} msec".format(average_time))
return
def processQRCodeDetection(self, qrcode, frame):
if len(frame.shape) == 2:
result = cv.cvtColor(frame, cv.COLOR_GRAY2BGR)
else:
result = frame
print('Run {:s} on video frame [{:d}x{:d}]'.format(
self.getQRModeString(), frame.shape[1], frame.shape[0]))
timer = cv.TickMeter()
timer.start()
points, decode_info = self.runQR(qrcode, frame)
timer.stop()
fps = 1 / timer.getTimeSec()
self.drawQRCodeResults(result, points, decode_info, fps)
return fps, result, points
def __init__(self):
# Load the parameters
self.conf = config()
# initialize dlib's face detector (HOG-based) and then create the
# facial landmark predictor
print("[INFO] loading facial landmark predictor...")
self.detector = dlib.get_frontal_face_detector()
self.predictor = dlib.shape_predictor(self.conf.shape_predictor_path)
# grab the indexes of the facial landmarks for the left and
# right eye, respectively
(self.lStart, self.lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]
(self.rStart, self.rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]
# initialize the video stream and sleep for a bit, allowing the
# camera sensor to warm up
self.cap = cv2.VideoCapture(0)
if self.conf.vedio_path == 0:
self.cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = self.cap.read()
# Introduce mark_detector to detect landmarks.
self.mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
self.img_queue = Queue()
self.box_queue = Queue()
self.img_queue.put(sample_frame)
self.box_process = Process(target=get_face, args=(
self.mark_detector, self.img_queue, self.box_queue,))
self.box_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
self.height, self.width = sample_frame.shape[:2]
self.pose_estimator = PoseEstimator(img_size=(self.height, self.width))
# Introduce scalar stabilizers for pose.
self.pose_stabilizers = [Stabilizer(
state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)]
self.tm = cv2.TickMeter()
# Gaze tracking
self.gaze = GazeTracking()
def init():
"""MAIN"""
# Video source from webcam or video file.
video_src = args.cam if args.cam is not None else args.video
if video_src is None:
print(
"Warning: video source not assigned, default webcam will be used.")
video_src = 0
cap = cv2.VideoCapture(video_src)
if video_src == 0:
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = cap.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
box_process = Process(target=get_face,
args=(
mark_detector,
img_queue,
box_queue,
))
box_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
return cap, video_src, img_queue, box_queue, tm, mark_detector, pose_estimator
def dnn(frame):
'''
Applies googlenet onto a window
'''
#print(frame)
img = cv2.resize(frame, (224,224), interpolation = cv2.INTER_CUBIC)
if(img is None):
return
inputBlob = cv2.dnn.blobFromImage(img, 1.0, (224,224),(104,117,123),False)
t = cv2.TickMeter()
t.start()
net.setInput(inputBlob, "data")
prob = net.forward("prob")
t.stop()
classId,_ = getMaxClass(prob)
classNames = readClassNames()
print(classNames[classId].rstrip())
return classNames[classId].rstrip()
def detect_video(video_file):
cap = cv2.VideoCapture(video_file)
assert cap.isOpened()
face_detector = pydetector.FaceDetector(model_path='./models', num_thread=1, scale=0.25)
meter = cv2.TickMeter()
while True:
ret, frame = cap.read()
if not ret:
break
# frame = cv2.rotate(frame, cv2.ROTATE_90_COUNTERCLOCKWISE)
meter.reset()
meter.start()
boxes = face_detector.detect(img_bgr=frame)
meter.stop()
for item in boxes:
cv2.rectangle(frame, (item.x, item.y), (item.x + item.width, item.y + item.height), (0, 255, 255), 2)
cv2.imshow('face-detect', frame)
print('time={:.3}ms'.format(meter.getTimeMilli()))
cv2.waitKeyEx(33)
def on_current_plugin_update_needed(self):
print("on_current_plugin_update_needed called")
if not (self.originalMat is None):
print("originalMat exists")
self.processedMat = self.originalMat.copy()
# else:
if not self.currentPluginGui is None:
print("starting to time process")
print(type(self.currentPlugin))
meter = cv.TickMeter()
meter.start()
self.processedMat = self.currentPluginGui.process_image(
self.originalMat, self.processedMat)
meter.stop()
print("The process took ", meter.getTimeMilli(), " milliseconds")
# cv.imshow("processed:", self.processedMat)
# cv.imshow("original:", self.originalMat)
# cv.waitKey(0); cv.destroyAllWindows()
self.originalImage = QImage(self.originalMat.data,
self.originalMat.shape[1],
self.originalMat.shape[0],
QImage.Format_RGB888)
temp_original_pixmap = QPixmap.fromImage(
self.originalImage.rgbSwapped())
self.originalPixmap.setPixmap(temp_original_pixmap)
# self.originalPixmap.fromImage(self.originalImage.rgbSwapped())
print(type(self.originalPixmap))
self.processedImage = QImage(self.processedMat.data,
self.processedMat.shape[1],
self.processedMat.shape[0],
QImage.Format_RGB888)
temp_processed_pixmap = QPixmap.fromImage(
self.processedImage.rgbSwapped())
self.processedPixmap.setPixmap(temp_processed_pixmap)
print("last line completed")
return
def main():
backends = (cv.dnn.DNN_BACKEND_DEFAULT,
cv.dnn.DNN_BACKEND_HALIDE,
cv.dnn.DNN_BACKEND_INFERENCE_ENGINE,
cv.dnn.DNN_BACKEND_OPENCV)
targets = (cv.dnn.DNN_TARGET_CPU,
cv.dnn.DNN_TARGET_OPENCL,
cv.dnn.DNN_TARGET_OPENCL_FP16,
cv.dnn.DNN_TARGET_MYRIAD)
parser = argparse.ArgumentParser(description='A demo for running libfacedetection using OpenCV\'s DNN module.')
parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DEFAULT, type=int,
help='Choose one of computation backends: '
'%d: automatically (by default), '
'%d: Halide language (http://halide-lang.org/), '
'%d: Intel\'s Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit), '
'%d: OpenCV implementation' % backends)
parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int,
help='Choose one of target computation devices: '
'%d: CPU target (by default), '
'%d: OpenCL, '
'%d: OpenCL fp16 (half-float precision), '
'%d: VPU' % targets)
# Location
parser.add_argument('--input', '-i', help='Path to the image. Omit to call default camera')
parser.add_argument('--model', '-m', type=str, help='Path to .onnx model file.')
# Inference parameters
parser.add_argument('--score_threshold', default=0.6, type=float, help='Threshold for filtering out faces with conf < conf_thresh.')
parser.add_argument('--nms_threshold', default=0.3, type=float, help='Threshold for non-max suppression.')
parser.add_argument('--top_k', default=5000, type=int, help='Keep keep_top_k for results outputing.')
# Result
parser.add_argument('--vis', default=True, type=str2bool, help='Set True to visualize the result image. Invalid when using camera.')
parser.add_argument('--save', default=False, type=str2bool, help='Set True to save as result.jpg. Invalid when using camera.')
args = parser.parse_args()
# Instantiate yunet
yunet = cv.FaceDetectorYN.create(
model=args.model,
config='',
input_size=(320, 320),
score_threshold=args.score_threshold,
nms_threshold=args.nms_threshold,
top_k=5000,
backend_id=args.backend,
target_id=args.target
)
if args.input is not None:
image = cv.imread(args.input)
yunet.setInputSize((image.shape[1], image.shape[0]))
_, faces = yunet.detect(image) # faces: None, or nx15 np.array
vis_image = visualize(image, faces)
if args.save:
print('result.jpg saved.')
cv.imwrite('result.jpg', vis_image)
if args.vis:
cv.namedWindow(args.input, cv.WINDOW_AUTOSIZE)
cv.imshow(args.input, vis_image)
cv.waitKey(0)
else:
device_id = 0
cap = cv.VideoCapture(device_id)
frame_w = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
frame_h = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
yunet.setInputSize([frame_w, frame_h])
tm = cv.TickMeter()
while cv.waitKey(1) < 0:
has_frame, frame = cap.read()
if not has_frame:
print('No frames grabbed!')
tm.start()
_, faces = yunet.detect(frame) # # faces: None, or nx15 np.array
tm.stop()
frame = visualize(frame, faces, fps=tm.getFPS())
cv.imshow('libfacedetection demo', frame)
tm.reset()
def main():
"""MAIN"""
# Video source from webcam or video file.
video_src = args.cam if args.cam is not None else args.video
if video_src is None:
print(
"Warning: video source not assigned, default webcam will be used.")
video_src = 0
cap = cv2.VideoCapture(video_src)
if video_src == 0:
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = cap.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
box_process = Process(target=get_face,
args=(
mark_detector,
img_queue,
box_queue,
))
box_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
# Uncomment to prepare for preview, define the codec and create VideoWriter object for output video
# fourcc = cv2.VideoWriter_fourcc(*'mp4v')
# out = cv2.VideoWriter('output.mp4', fourcc, 20.0, (width, height))
while True:
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = cap.read()
# Break if failure or video finished
if frame_got is False:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
if video_src == 0:
frame = cv2.flip(frame, 2)
# Pose estimation by 3 steps:
# 1. detect face;
# 2. detect landmarks;
# 3. estimate pose
# Feed frame to image queue.
img_queue.put(frame)
# Get face from box queue.
facebox = box_queue.get()
if facebox is not None:
# Detect landmarks from image of 128x128.
face_img = frame[facebox[1]:facebox[3], facebox[0]:facebox[2]]
face_img = cv2.resize(face_img, (CNN_INPUT_SIZE, CNN_INPUT_SIZE))
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
# Get marks
marks = mark_detector.detect_marks([face_img])
# Convert the marks locations from local CNN to global image.
marks *= (facebox[2] - facebox[0])
marks[:, 0] += facebox[0]
marks[:, 1] += facebox[1]
# Uncomment following line to show raw marks.
# mark_detector.draw_marks(
# frame, marks, color=(0, 255, 0))
# Try pose estimation with 68 points.
pose = pose_estimator.solve_pose_by_68_points(marks)
# Stabilize the pose.
steady_pose = []
pose_np = np.array(pose).flatten()
for value, ps_stb in zip(pose_np, pose_stabilizers):
ps_stb.update([value])
steady_pose.append(ps_stb.state[0])
steady_pose = np.reshape(steady_pose, (-1, 3))
# Uncomment following line to draw stable pose annotation on frame.
pose_estimator.draw_annotation_box(frame,
steady_pose[0],
steady_pose[1],
color=(128, 255, 128))
# Uncomment following line to draw head axes on frame.
pose_estimator.draw_axis(frame, steady_pose[0], steady_pose[1])
# Uncomment following line to get the length of the line on the face
# faced_line_length = pose_estimator.get_faced_line_length(frame, steady_pose[0], steady_pose[1])
# Print status "FACING"
# cv2.putText(frame, "FACING", (height//50, width//50), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 0, 0), 1)
print("FACING")
else:
# Print status "UNFACED"
# cv2.putText(frame, "UNFACED", (height//50, width//50), cv2.FONT_HERSHEY_DUPLEX, 1, (0, 0, 0), 1)
print("UNFACED")
# Uncomment to write frame
# out.write(frame)
# Uncommet to show preview.
# cv2.imshow("Preview", frame)
# Wait
if cv2.waitKey(10) == 27:
break
# Uncomment to release output file
# out.release()
# Clean up the multiprocessing process.
box_process.terminate()
box_process.join()
if video_src == 0:
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
# Get the real frame resolution.
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_rate = cap.get(cv2.CAP_PROP_FPS)
# Video output by video writer.
if args.write_video:
video_writer = cv2.VideoWriter('output.avi',
cv2.VideoWriter_fourcc(*'avc1'),
frame_rate, (frame_width, frame_height))
# Introduce a metter to measure the FPS.
tm_detection = cv2.TickMeter()
tm_identification = cv2.TickMeter()
# Loop through the video frames.
while True:
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = cap.read()
if not frame_got:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
if video_src == 0:
frame = cv2.flip(frame, 2)
def main():
# 图片路径
img_path = "./image/image.png"
# 算法名称 edsr, espcn, fsrcnn or lapsrn
algorithm = "lapsrn"
# 模型路径，根据算法确定
model = "./model/LapSRN_x2.pb"
# 放大系数
scale = 2
# 时间系数
perf = []
img = cv2.imread(img_path)
if img is None:
print("Couldn't load image: " + str(img_path))
# Crop the image so the images will be aligned
# 裁剪图像，使图像对齐
width = img.shape[0] - (img.shape[0] % scale)
height = img.shape[1] - (img.shape[1] % scale)
cropped = img[0:width, 0:height]
# Downscale the image for benchmarking
# 缩小图像，以实现基准质量测试
img_downscaled = cv2.resize(cropped, None, fx=1.0 / scale, fy=1.0 / scale)
# Make dnn super resolution instance
# 超分模型初始化
sr = dnn_superres.DnnSuperResImpl_create()
# Read and set the dnn model
# 读取和设定模型
sr.readModel(model)
sr.setModel(algorithm, scale)
timer = cv2.TickMeter()
timer.start()
# 放大图像
img_new = sr.upsample(img_downscaled)
timer.stop()
# 运行时间s
elapsed = timer.getTimeSec() / timer.getCounter()
perf.append(elapsed)
print(sr.getAlgorithm() + " : " + str(elapsed))
# INTER_CUBIC - 三次样条插值放大图像
timer.start()
bicubic = cv2.resize(img_downscaled,
None,
fx=scale,
fy=scale,
interpolation=cv2.INTER_CUBIC)
timer.stop()
# 运行时间s
elapsed = timer.getTimeSec() / timer.getCounter()
perf.append(elapsed)
print("Bicubic" + " : " + str(elapsed))
# INTER_NEAREST - 最近邻插值
timer.start()
nearest = cv2.resize(img_downscaled,
None,
fx=scale,
fy=scale,
interpolation=cv2.INTER_NEAREST)
timer.stop()
# 运行时间s
elapsed = timer.getTimeSec() / timer.getCounter()
perf.append(elapsed)
print("Nearest" + " : " + str(elapsed))
# Lanczos插值放大图像
timer.start()
lanczos = cv2.resize(img_downscaled,
None,
fx=scale,
fy=scale,
interpolation=cv2.INTER_LANCZOS4)
timer.stop()
# 运行时间s
elapsed = timer.getTimeSec() / timer.getCounter()
perf.append(elapsed)
print("Lanczos" + " : " + str(elapsed))
imgs = [img_new, bicubic, nearest, lanczos]
titles = [sr.getAlgorithm(), "Bicubic", "Nearest neighbor", "Lanczos"]
showBenchmark(imgs, titles, perf)
def main():
# Read and store arguments
confThreshold = args.thr
nmsThreshold = args.nms
inpWidth = args.width
inpHeight = args.height
modelDetector = args.model
modelRecognition = args.ocr
# Load network
detector = cv.dnn.readNet(modelDetector)
recognizer = cv.dnn.readNet(modelRecognition)
# Create a new named window
kWinName = "EAST: An Efficient and Accurate Scene Text Detector"
cv.namedWindow(kWinName, cv.WINDOW_NORMAL)
outNames = []
outNames.append("feature_fusion/Conv_7/Sigmoid")
outNames.append("feature_fusion/concat_3")
# Open a video file or an image file or a camera stream
cap = cv.VideoCapture(args.input if args.input else 0)
tickmeter = cv.TickMeter()
while cv.waitKey(1) < 0:
# Read frame
hasFrame, frame = cap.read()
if not hasFrame:
cv.waitKey()
break
# Get frame height and width
height_ = frame.shape[0]
width_ = frame.shape[1]
rW = width_ / float(inpWidth)
rH = height_ / float(inpHeight)
# Create a 4D blob from frame.
blob = cv.dnn.blobFromImage(frame, 1.0, (inpWidth, inpHeight), (123.68, 116.78, 103.94), True, False)
# Run the detection model
detector.setInput(blob)
tickmeter.start()
outs = detector.forward(outNames)
tickmeter.stop()
# Get scores and geometry
scores = outs[0]
geometry = outs[1]
[boxes, confidences] = decodeBoundingBoxes(scores, geometry, confThreshold)
# Apply NMS
indices = cv.dnn.NMSBoxesRotated(boxes, confidences, confThreshold, nmsThreshold)
for i in indices:
# get 4 corners of the rotated rect
vertices = cv.boxPoints(boxes[i[0]])
# scale the bounding box coordinates based on the respective ratios
for j in range(4):
vertices[j][0] *= rW
vertices[j][1] *= rH
# get cropped image using perspective transform
if modelRecognition:
cropped = fourPointsTransform(frame, vertices)
cropped = cv.cvtColor(cropped, cv.COLOR_BGR2GRAY)
# Create a 4D blob from cropped image
blob = cv.dnn.blobFromImage(cropped, size=(100, 32), mean=127.5, scalefactor=1 / 127.5)
recognizer.setInput(blob)
# Run the recognition model
tickmeter.start()
result = recognizer.forward()
tickmeter.stop()
# decode the result into text
wordRecognized = decodeText(result)
cv.putText(frame, wordRecognized, (int(vertices[1][0]), int(vertices[1][1])), cv.FONT_HERSHEY_SIMPLEX,
0.5, (255, 0, 0))
for j in range(4):
p1 = (vertices[j][0], vertices[j][1])
p2 = (vertices[(j + 1) % 4][0], vertices[(j + 1) % 4][1])
cv.line(frame, p1, p2, (0, 255, 0), 1)
# Put efficiency information
label = 'Inference time: %.2f ms' % (tickmeter.getTimeMilli())
cv.putText(frame, label, (0, 15), cv.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0))
# Display the frame
cv.imshow(kWinName, frame)
tickmeter.reset()
def main():
#bagreader = BagFileReader(args.video, 640,480,848,480,30,30)
bagreader = BagFileReader(args.video, 640, 480, 640, 480, 15, 15)
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
sample_frame = bagreader.get_color_frame()
sample_frame = cv2.cvtColor(sample_frame, cv2.COLOR_BGR2RGB)
height, width, _ = sample_frame.shape
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
out = cv2.VideoWriter('output-%s.avi' % args.name_output, fourcc, args.fps,
(width, height))
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
box_process = Process(target=get_face,
args=(
mark_detector,
img_queue,
box_queue,
))
box_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
while True:
t1 = time.time()
# Read frame, crop it, flip it, suits your needs.
frame = bagreader.get_color_frame()
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if frame is False:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
# if video_src == 0:
# frame = cv2.flip(frame, 2)
# Pose estimation by 3 steps:
# 1. detect face;
# 2. detect landmarks;
# 3. estimate pose
# Feed frame to image queue.
img_queue.put(frame)
# Get face from box queue.
faceboxes = box_queue.get()
print(faceboxes)
mess = "Not detect pose"
if faceboxes is not None:
if isinstance(faceboxes[1], int):
faceboxes = [faceboxes]
for facebox in faceboxes:
# Detect landmarks from image of 128x128.
face_img = frame[facebox[1]:facebox[3], facebox[0]:facebox[2]]
face_img = cv2.resize(face_img,
(CNN_INPUT_SIZE, CNN_INPUT_SIZE))
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
tm.start()
marks = mark_detector.detect_marks([face_img])
tm.stop()
# Convert the marks locations from local CNN to global image.
marks *= (facebox[2] - facebox[0])
marks[:, 0] += facebox[0]
marks[:, 1] += facebox[1]
# Uncomment following line to show raw marks.
# mark_detector.draw_marks(
# frame, marks, color=(0, 255, 0))
# Uncomment following line to show facebox.
# mark_detector.draw_box(frame, [facebox])
# Try pose estimation with 68 points.
pose = pose_estimator.solve_pose_by_68_points(marks)
# Stabilize the pose.
steady_pose = []
pose_np = np.array(pose).flatten()
for value, ps_stb in zip(pose_np, pose_stabilizers):
ps_stb.update([value])
steady_pose.append(ps_stb.state[0])
steady_pose = np.reshape(steady_pose, (-1, 3))
# Uncomment following line to draw pose annotation on frame.
pose_estimator.draw_annotation_box(frame,
pose[0],
pose[1],
color=(255, 128, 128))
# Uncomment following line to draw stabile pose annotation on frame.
t2 = time.time()
mess = round(1 / (t2 - t1), 2)
# pose_estimator.draw_annotation_box(
# frame, steady_pose[0], steady_pose[1], color=(128, 255, 128))
# Uncomment following line to draw head axes on frame.
# pose_estimator.draw_axes(frame, stabile_pose[0], stabile_pose[1])
cv2.putText(frame,
"FPS: " + "{}".format(mess), (20, 20),
cv2.FONT_HERSHEY_SIMPLEX,
0.75, (0, 255, 0),
thickness=2)
# Show preview.
cv2.imshow("Preview", frame)
out.write(frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
out.release()
# Clean up the multiprocessing process.
box_process.terminate()
box_process.join()
def main(strargument):
shutil.rmtree("./test")
os.mkdir("./test")
os.remove("result.txt")
f = open("result.txt", "a")
cap = cv2.VideoCapture(strargument)
# cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
#cap = cv2.VideoCapture("NTQ.mkv")
#cap = cv2.VideoCapture("/home/fitmta/Real-Time-Face-Detection-OpenCV-GPU/videos/video/out1.1.avi")
#cap = cv2.VideoCapture("http://*****:*****@#[email protected]:8932/mjpg/video.mjpg")
# cap = cv2.VideoCapture("http://*****:*****@[email protected]:8933/Streaming/channels/102/preview")
success, frame = cap.read()
startId = countIdFolder("./face_db/")
# quit if unable to read the video file
if not success:
print('Failed to read video')
sys.exit(1)
#The color of the rectangle we draw around the face
rectangleColor = (0, 165, 255)
#variables holding the current frame number and the current faceid
frameCounter = 0
currentFaceID = 0
#Variables holding the correlation trackers and the name per faceid
conf = configparser.ConfigParser()
conf.read("config/main.cfg")
mtcnn_detector = load_mtcnn(conf)
MODEL_PATH = conf.get("MOBILEFACENET", "MODEL_PATH")
VERIFICATION_THRESHOLD = float(
conf.get("MOBILEFACENET", "VERIFICATION_THRESHOLD"))
FACE_DB_PATH = conf.get("MOBILEFACENET", "FACE_DB_PATH")
BLUR_THRESH = int(conf.get("CUSTOM", "BLUR_THRESH"))
MIN_FACE_SIZE = int(conf.get("MTCNN", "MIN_FACE_SIZE"))
MAX_BLACK_PIXEL = int(conf.get("CUSTOM", "MAX_BLACK_PIXEL"))
YAWL = int(conf.get("CUSTOM", "YAWL"))
YAWR = int(conf.get("CUSTOM", "YAWR"))
PITCHL = int(conf.get("CUSTOM", "PITCHL"))
PITCHR = int(conf.get("CUSTOM", "PITCHR"))
ROLLL = int(conf.get("CUSTOM", "ROLLL"))
ROLLR = int(conf.get("CUSTOM", "ROLLR"))
MAXDISAPPEARED = int(conf.get("CUSTOM", "MAXDISAPPEARED"))
IS_FACE_THRESH = float(conf.get("CUSTOM", "IS_FACE_THRESH"))
EXTEND_Y = int(conf.get("CUSTOM", "EXTEND_Y"))
EXTEND_X = int(conf.get("CUSTOM", "EXTEND_X"))
SIMILAR_THRESH = float(conf.get("CUSTOM", "SIMILAR_THRESH"))
MAX_LIST_LEN = int(conf.get("CUSTOM", "MAX_LIST_LEN"))
MIN_FACE_FOR_SAVE = int(conf.get("CUSTOM", "MIN_FACE_FOR_SAVE"))
LIVE_TIME = int(conf.get("CUSTOM", "LIVE_TIME"))
ROIXL = int(conf.get("CUSTOM", "ROIXL"))
ROIXR = int(conf.get("CUSTOM", "ROIXR"))
ROIYB = int(conf.get("CUSTOM", "ROIYB"))
ROIYA = int(conf.get("CUSTOM", "ROIYA"))
maxDisappeared = MAXDISAPPEARED ## khong xuat hien toi da 100 frame
faces_db = load_faces(FACE_DB_PATH, mtcnn_detector)
# load_face_db = ThreadingUpdatefacedb(FACE_DB_PATH,mtcnn_detector)
time.sleep(10)
for item in faces_db:
print(item["name"])
listTrackedFace = []
mark_detector = MarkDetector()
tm = cv2.TickMeter()
_, sample_frame = cap.read()
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
with tf.Graph().as_default():
with tf.Session() as sess:
load_mobilefacenet(MODEL_PATH)
inputs_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
try:
start = time.time()
while True:
start1 = time.time()
retval, frame = cap.read()
#Increase the framecounter
frameCounter += 1
if retval:
_frame = frame[ROIYA:ROIYB, ROIXL:ROIXR]
cv2.rectangle(frame, (ROIXL, ROIYA), (ROIXR, ROIYB),
(0, 0, 255), 2)
good_face_index = []
# faces_db = load_face_db.face_db
if (frameCounter % 1) == 0:
### embed and compare name
for i, face_db in enumerate(faces_db):
if not os.path.isdir(
"./face_db/" +
face_db["name"].split("_")[0]):
faces_db.pop(i)
faces, landmarks = mtcnn_detector.detect(_frame)
if faces.shape[0] is not 0:
input_images = np.zeros(
(faces.shape[0], 112, 112, 3))
save_images = np.zeros(
(faces.shape[0], 112, 112, 3))
(yaw, pitch, roll) = (0, 0, 0)
for i, face in enumerate(faces):
if round(faces[i, 4], 6) > IS_FACE_THRESH:
bbox = faces[i, 0:4]
points = landmarks[i, :].reshape(
(5, 2))
nimg = face_preprocess.preprocess(
_frame,
bbox,
points,
image_size='112,112')
save_images[i, :] = nimg
nimg = nimg - 127.5
nimg = nimg * 0.0078125
input_images[i, :] = nimg
(x1, y1, x2, y2) = bbox.astype("int")
if x1 < 0 or y1 < 0 or x2 < 0 or y2 < 0 or x1 >= x2 or y1 >= y2:
continue
temp = int((y2 - y1) / EXTEND_Y)
y1 = y1 + temp
y2 = y2 + temp
temp = int((x2 - x1) / EXTEND_X)
if x1 > temp:
x1 = x1 - temp
x2 = x2 + temp
# cv2.imshow("mainframe",frame)
# cv2.imwrite("temp2.jpg",frame[y1:y2,x1:x2])
face_img = cv2.resize(
_frame[y1:y2, x1:x2], (128, 128))
# cv2.imshow("ok",face_img)
face_img = cv2.cvtColor(
face_img, cv2.COLOR_BGR2RGB)
tm.start()
marks = mark_detector.detect_marks(
[face_img])
tm.stop()
marks *= (x2 - x1)
marks[:, 0] += x1
marks[:, 1] += y1
# mark_detector.draw_marks(
# frame, marks, color=(0, 255, 0))
pose, (
yaw, pitch, roll
) = pose_estimator.solve_pose_by_68_points(
marks)
# temp = frame
# cv2.putText(temp,"yaw: "+str(yaw),(x2,y1),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 255), thickness=2)
# cv2.putText(temp,"pitch: "+str(pitch),(x2,y1+25),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 255), thickness=2)
# cv2.putText(temp,"roll: "+str(roll),(x2,y1+50),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 255), thickness=2)
# cv2.imshow("frame",temp)
# if measureGoodFace(MIN_FACE_SIZE,MAX_BLACK_PIXEL,frame[y1:y2,x1:x2],yaw,pitch,roll,BLUR_THRESH,YAWL,YAWR,PITCHL,PITCHR,ROLLL,ROLLR):
# good_face_index.append(i)
# cv2.waitKey(0)
# print(good_face_index)
feed_dict = {inputs_placeholder: input_images}
emb_arrays = sess.run(embeddings,
feed_dict=feed_dict)
emb_arrays = sklearn.preprocessing.normalize(
emb_arrays)
names = []
sims = []
for i, embedding in enumerate(emb_arrays):
# if len(listTrackedFace)>i and RepresentsInt(listTrackedFace[i].name)==False:
# names.append(listTrackedFace[i].name)
# continue
embedding = embedding.flatten()
temp_dict = {}
for com_face in faces_db:
ret, sim = feature_compare(
embedding, com_face["feature"],
0.65)
temp_dict[com_face["name"]] = sim
# print(temp_dict)
dictResult = sorted(temp_dict.items(),
key=lambda d: d[1],
reverse=True)
# print(dictResult[:5])
name = ""
if len(dictResult) > 0 and dictResult[0][
1] > VERIFICATION_THRESHOLD:
name = dictResult[0][
0] #.split("_")[0]
sim = dictResult[0][1]
## wite log
t = time.time()
f.write(name + "___" +
str((t - start) // 60) + ":" +
str(int(t - start) % 60) +
"\n")
else:
name = "unknown"
sim = 0
names.append(name)
sims.append(sim)
# cv2.imwrite("./test/"+name+"_"+str(frameCounter//60)+":"+str(frameCounter%60)+".jpg",save_images[i,:])
# if len(dictResult)>0 :
# cv2.imwrite("./test/"+names[i]+"_"+str(frameCounter//60)+":"+str(frameCounter%60)+"_"+str(dictResult[0][1])+".jpg",save_images[i,:])
################################ tracker
for i, embedding in enumerate(emb_arrays):
embedding = embedding.flatten()
ResultDict = {}
for objectTrackFace in listTrackedFace:
tempList = []
(x1, y1, x2,
y2) = objectTrackFace.latestBox
for com_face in objectTrackFace.listEmbedding:
ret, sim = feature_compare(
embedding, com_face, 0.65)
tempList.append(sim)
tempList.sort(reverse=True)
if len(tempList) > 0:
if tempList[0] > 0.9 or (
similarIOU(
faces[i, :4].astype(
"int"),
objectTrackFace.
latestBox) and
(frameCounter - objectTrackFace
.latestFrameCounter) < 3):
ResultDict[objectTrackFace.
name] = tempList[0]
dictResult = sorted(ResultDict.items(),
key=lambda d: d[1],
reverse=True)
if True:
if len(
ResultDict
) > 0 and dictResult[0][
1] > SIMILAR_THRESH: ## neu khop -- 0.5
# for ik in range(len(dict)):
# if dict[ik][1]>SIMILAR_THRESH:
nameTrackCurrent = dictResult[0][0]
for index, tempFaceTrack in enumerate(
listTrackedFace):
if tempFaceTrack.name == nameTrackCurrent:
if len(tempFaceTrack.
listImage
) > MAX_LIST_LEN:
tempFaceTrack.listImage.pop(
0)
tempFaceTrack.listEmbedding.pop(
0)
if measureGoodFace(
MIN_FACE_SIZE,
MAX_BLACK_PIXEL,
save_images[
i, :], yaw,
pitch, roll,
BLUR_THRESH,
YAWL, YAWR,
PITCHL, PITCHR,
ROLLL, ROLLR):
tempFaceTrack.listImage.append(
save_images[
i, :])
tempFaceTrack.listEmbedding.append(
emb_arrays[i])
else:
if measureGoodFace(
MIN_FACE_SIZE,
MAX_BLACK_PIXEL,
save_images[
i, :], yaw,
pitch, roll,
BLUR_THRESH,
YAWL, YAWR,
PITCHL, PITCHR,
ROLLL, ROLLR):
tempFaceTrack.listImage.append(
save_images[
i, :])
tempFaceTrack.listEmbedding.append(
emb_arrays[i])
if names[i] != "unknown":
if RepresentsInt(
nameTrackCurrent
):
tempFaceTrack.name = names[
i]
# else: #################
# names[i] = nameTrackCurrent
else:
if not RepresentsInt(
nameTrackCurrent
):
names[
i] = nameTrackCurrent
tempFaceTrack.countDisappeared = 0
tempFaceTrack.latestBox = faces[
i, 0:4].astype("int")
tempFaceTrack.latestFrameCounter = frameCounter
tempFaceTrack.liveTime = 0
tempFaceTrack.justAdded = True ## but we still action with it
break
else: ## neu khong khop thi tao moi nhung chi them anh khi mat du tot
if len(ResultDict) > 0:
print(dictResult[0][1])
if names[i] != "unknown":
newTrackFace = trackedFace(
names[i])
else:
newTrackFace = trackedFace(
str(currentFaceID))
currentFaceID = currentFaceID + 1
if measureGoodFace(
MIN_FACE_SIZE,
MAX_BLACK_PIXEL,
save_images[i, :], yaw,
pitch, roll, BLUR_THRESH,
YAWL, YAWR, PITCHL, PITCHR,
ROLLL, ROLLR):
newTrackFace.listImage.append(
save_images[i, :])
newTrackFace.listEmbedding.append(
emb_arrays[i])
newTrackFace.latestBox = faces[
i, 0:4].astype("int")
newTrackFace.latestFrameCounter = frameCounter
# print(newTrackFace.latestBox)
newTrackFace.justAdded = True
listTrackedFace.append(
newTrackFace) ## add list
### disappeared
for index, trackFace in enumerate(
listTrackedFace):
if trackFace.justAdded == False:
trackFace.countDisappeared = trackFace.countDisappeared + 1
trackFace.liveTime = trackFace.liveTime + 1
else:
trackFace.justAdded = False
if trackFace.liveTime > LIVE_TIME:
t = listTrackedFace.pop(index)
del t
if trackFace.countDisappeared > maxDisappeared:
if len(
trackFace.listImage
) < MIN_FACE_FOR_SAVE: ## neu chua duoc it nhat 5 mat thi xoa luon
trackedFace.countDisappeared = 0
continue
if trackFace.saveTrackedFace(
"./temp/", startId):
startId = startId + 1
t = listTrackedFace.pop(index)
del t
for i, face in enumerate(faces):
x1, y1, x2, y2 = faces[i][0], faces[i][
1], faces[i][2], faces[i][3]
x1 = max(int(x1), 0)
y1 = max(int(y1), 0)
x2 = min(int(x2), _frame.shape[1])
y2 = min(int(y2), _frame.shape[0])
cv2.rectangle(frame,
(x1 + ROIXL, y1 + ROIYA),
(x2 + ROIXL, y2 + ROIYA),
(0, 255, 0), 2)
# if i in good_face_index:
# if not RepresentsInt(names[i]):
cv2.putText(frame, names[i].split("_")[0],
(int(x1 / 2 + x2 / 2 + ROIXL),
int(y1 + ROIYA)),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 255, 255), 2)
else:
for index, trackFace in enumerate(
listTrackedFace):
trackFace.countDisappeared = trackFace.countDisappeared + 1
trackFace.liveTime = trackFace.liveTime + 1
if trackFace.liveTime > LIVE_TIME:
t = listTrackedFace.pop(index)
del t
continue
if trackFace.countDisappeared > maxDisappeared:
if len(
trackFace.listImage
) < MIN_FACE_FOR_SAVE: ## neu chua duoc it nhat 5 mat thi xoa luon
trackedFace.countDisappeared = 0
continue
if trackFace.saveTrackedFace(
"./temp/", startId):
startId = startId + 1
t = listTrackedFace.pop(index)
del t
end = time.time()
cv2.putText(frame,
"FPS: " + str(1 // (end - start1)),
(400, 30), cv2.FONT_HERSHEY_SIMPLEX, 1,
(0, 0, 0), 3)
cv2.putText(
frame, "Time: " + str((end - start) // 60) +
":" + str(int(end - start) % 60), (200, 30),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 3)
cv2.imshow("frame", frame)
key = cv2.waitKey(30)
if key & 0xFF == ord('q'):
break
if key == 32:
cv2.waitKey(0)
else:
break
except KeyboardInterrupt as e:
pass
gc.collect()
cv2.destroyAllWindows()
exit(0)
def main():
"""MAIN"""
# Video source from webcam or video file.
video_src = 0
#video_src = "./head-pose-face-detection-female.mp4"
cam = cv2.VideoCapture(video_src)
if video_src == 0:
cam.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = cam.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
thread = threading.Thread(target=get_face,
args=(mark_detector, img_queue, box_queue))
thread.daemon = True
thread.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
while True:
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = cam.read()
if frame_got is False:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
if video_src == 0:
frame = cv2.flip(frame, 2)
# Pose estimation by 3 steps:
# 1. detect face;
# 2. detect landmarks;
# 3. estimate pose
# Feed frame to image queue.
img_queue.put(frame)
# Get face from box queue.
facebox = box_queue.get()
if facebox is not None:
# Detect landmarks from image of 128x128.
face_img = frame[facebox[1]:facebox[3], facebox[0]:facebox[2]]
face_img = cv2.resize(face_img, (CNN_INPUT_SIZE, CNN_INPUT_SIZE))
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
tm.start()
marks = mark_detector.detect_marks(face_img)
tm.stop()
#print(tm.getTimeSec()/tm.count())
# Convert the marks locations from local CNN to global image.
marks *= (facebox[2] - facebox[0])
marks[:, 0] += facebox[0]
marks[:, 1] += facebox[1]
# Uncomment following line to show raw marks.
# mark_detector.draw_marks(
# frame, marks, color=(0, 255, 0))
# Try pose estimation with 68 points.
pose = pose_estimator.solve_pose_by_68_points(marks)
# Stabilize the pose.
stabile_pose = []
pose_np = np.array(pose).flatten()
for value, ps_stb in zip(pose_np, pose_stabilizers):
ps_stb.update([value])
stabile_pose.append(ps_stb.state[0])
stabile_pose = np.reshape(stabile_pose, (-1, 3))
# Uncomment following line to draw pose annotation on frame.
# pose_estimator.draw_annotation_box(
# frame, pose[0], pose[1], color=(255, 128, 128))
# Uncomment following line to draw stabile pose annotation on frame.
pose_estimator.draw_annotation_box(frame,
stabile_pose[0],
stabile_pose[1],
color=(128, 255, 128))
# Show preview.
cv2.imshow("Preview", frame)
if cv2.waitKey(10) == 27:
break
def process_video_capture(
video_capture,
save_file,
inverter_filename='models/inverter/inverter_randforest_7000.pkl'):
Z_DIM = 512
CNN_INPUT_SIZE = 128
sess = tf.InteractiveSession()
video_capture.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
# Check success
if not video_capture.isOpened():
raise Exception("Could not open video device")
_, sample_frame = video_capture.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
G, I = load_models(inverter_filename=inverter_filename)
#print(I.coef_)
#print(I.intercept_)
age_net = cv2.dnn.readNetFromCaffe('models/race_age/deploy_age.prototxt',
'models/race_age/age_net.caffemodel')
gender_net = cv2.dnn.readNetFromCaffe(
'models/race_age/deploy_gender.prototxt',
'models/race_age/gender_net.caffemodel')
i = 0
black_img = np.zeros((1024, 1024, 3))
while True:
#I.intercept_ += np.random.normal(scale=0.1, size=I.intercept_.shape).astype(np.float32)
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = video_capture.read()
if frame_got is False:
break
# Crop it if frame is larger than expected.
#frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
#frame = cv2.flip(frame, 2)
# Feed frame to image queue.
facebox = mark_detector.extract_cnn_facebox(frame)
if facebox is not None:
feats = calculate_facial_features(frame, facebox, tm,
CNN_INPUT_SIZE, mark_detector,
age_net, gender_net)
viewer_latent = I.predict(feats.reshape(1, -1))
labels = np.zeros([viewer_latent.shape[0], 0], np.float32)
viewer_generated = G.run(viewer_latent,
labels,
out_mul=127.5,
out_add=127.5,
out_dtype=np.uint8)
viewer_generated = np.squeeze(viewer_generated)
viewer_generated = np.transpose(viewer_generated, (1, 2, 0))
viewer_generated = cv2.cvtColor(viewer_generated,
cv2.COLOR_BGR2RGB)
facebox = mark_detector.extract_cnn_facebox(viewer_generated)
if facebox is not None:
feats_gen = calculate_facial_features(viewer_generated,
facebox, tm,
CNN_INPUT_SIZE,
mark_detector, age_net,
gender_net)
print("Frame: {}, mean diff in features: {} ".format(
i, np.mean(np.abs(feats_gen - feats))))
# Show preview.
else:
print("Frame: {}, didn't generate face".format(i))
viewer_generated = black_img
frame = cv2.resize(frame, (1024, 1024))
result_both_imgs = np.concatenate((viewer_generated, frame), axis=0)
cv2.imwrite("images/output_{}.png".format(i), result_both_imgs)
#last_image = viewer_generated
#cv2.imshow("Preview", viewer_generated)
if cv2.waitKey(10) == 27:
break
i += 1
# Clean up the multiprocessing process.
video_capture.release()
frames_to_video.create_video_from_frames(dir_path='images',
ext='png',
output=save_file)
def ai_obj_det(capd_img_list):
# Input Image Size into AI model
IN_WIDTH = 300 # have to fix 300x300
IN_HEIGHT = 300
frame_width = 640
frame_height = 480
# Defined Label list of Studied Model (Mobilenet SSD COCO)
CLASS_LABELS = {
0: 'background',
1: 'person',
2: 'bicycle',
3: 'car',
4: 'motorcycle',
5: 'airplane',
6: 'bus',
7: 'train',
8: 'truck',
9: 'boat',
10: 'traffic light',
11: 'fire hydrant',
13: 'stop sign',
14: 'parking meter',
15: 'bench',
16: 'bird',
17: 'cat',
18: 'dog',
19: 'horse',
20: 'sheep',
21: 'cow',
22: 'elephant',
23: 'bear',
24: 'zebra',
25: 'giraffe',
27: 'backpack',
28: 'umbrella',
31: 'handbag',
32: 'tie',
33: 'suitcase',
34: 'frisbee',
35: 'skis',
36: 'snowboard',
37: 'sports ball',
38: 'kite',
39: 'baseball bat',
40: 'baseball glove',
41: 'skateboard',
42: 'surfboard',
43: 'tennis racket',
44: 'bottle',
46: 'wine glass',
47: 'cup',
48: 'fork',
49: 'knife',
50: 'spoon',
51: 'bowl',
52: 'banana',
53: 'apple',
54: 'sandwich',
55: 'orange',
56: 'broccoli',
57: 'carrot',
58: 'hot dog',
59: 'pizza',
60: 'donut',
61: 'cake',
62: 'chair',
63: 'couch',
64: 'potted plant',
65: 'bed',
67: 'dining table',
70: 'toilet',
72: 'tv',
73: 'laptop',
74: 'mouse',
75: 'remote',
76: 'keyboard',
77: 'cell phone',
78: 'microwave',
79: 'oven',
80: 'toaster',
81: 'sink',
82: 'refrigerator',
84: 'book',
85: 'clock',
86: 'vase',
87: 'scissors',
88: 'teddy bear',
89: 'hair drier',
90: 'toothbrush'
}
# Define "Argument(Set Command Option)"
ap = argparse.ArgumentParser()
ap.add_argument('-p', '--pbtxt', required=True, help='path to pbtxt file')
ap.add_argument('-w',
'--weights',
required=True,
help='path to TensorFlow inference graph')
ap.add_argument('-c',
'--confidence',
type=float,
default=0.3,
help='minimum probability')
ap.add_argument('-i',
'--interval',
type=float,
default=0,
help='process interval to reduce CPU usage')
args = vars(ap.parse_args())
# Stop watch function from OpenCV as tm
# kW: may not necesary
tm = cv2.TickMeter()
colors = {}
# Random set on frame color for each label
random.seed()
for key in CLASS_LABELS.keys():
colors[key] = (random.randrange(255), random.randrange(255),
random.randrange(255))
# Loading AI model
print('Loading AI Model...')
net = cv2.dnn.readNet(args['weights'], args['pbtxt'])
for j in range(len(capd_img_list)):
image = cv2.imread(capd_img_list[j])
# convert to blob format from modified iputimage
blob = cv2.dnn.blobFromImage(image,
size=(IN_WIDTH, IN_HEIGHT),
swapRB=False,
crop=False)
# Set captured image into AImodel set as blob format
net.setInput(blob)
# Load image into AI
tm.reset()
tm.start()
detections = net.forward(
) # Load Image into AI model(net = cv2.dnn.readNet)
tm.stop()
# Repeat detection
for i in range(detections.shape[2]):
# pick up "i"th detected object's correct answer ratio
confidence = detections[0, 0, i, 2]
# will not do nothing, if lower than correct answer ratio
if confidence < args['confidence']:
continue
# Obtain Type and (x,y) of detected object
class_id = int(detections[0, 0, i, 1])
start_x = int(detections[0, 0, i, 3] * frame_width)
start_y = int(detections[0, 0, i, 4] * frame_height)
end_x = int(detections[0, 0, i, 5] * frame_width)
end_y = int(detections[0, 0, i, 6] * frame_height)
# Set the Name label like "peroson","cat" to detemin object name
label = CLASS_LABELS[class_id]
label += ': ' + str(round(confidence * 100, 2)) + '%'
label_size, base_line = cv2.getTextSize(label,
cv2.FONT_HERSHEY_SIMPLEX,
0.5, 1)
# Draw rectangle in the image
if CLASS_LABELS[class_id] == 'person': # Person or not
#cv2.rectangle(image, (start_x, start_y), (end_x, end_y), colors[class_id], -1) # thickness : -1 patinted
cv2.rectangle(image, (start_x, start_y), (end_x, end_y),
colors[class_id], 2) # thickness : -1 patinted
else:
cv2.rectangle(image, (start_x, start_y), (end_x, end_y),
colors[class_id], 2) # thickness : 2
# Debugg
print("Label: %s," % label, "Confidence: %f," % confidence)
# Draw Rectangle
cv2.rectangle(image, (start_x, start_y - label_size[1]),
(start_x + label_size[0], start_y + base_line),
(255, 255, 255), cv2.FILLED)
# Draw Text
cv2.putText(image, label, (start_x, start_y),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0))
# Measure Time and show on Text
ai_time = tm.getTimeMilli()
# Show AI Processing Time
#cv2.putText(image, '{:.2f}(ms)'.format(ai_time), (10,30), cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 255, 0), thickness=2)
'''
# Debug show image
cv2.imshow(capd_img_list[j],image)
cv2.waitKey(0)
cv2.destroyAllWindows()
'''
# Save AI overlaied image
cv2.imwrite(capd_img_list[j], image)
# Overlay
time.sleep(args['interval']) # Set inteval to repeat do AI
# Close Process
print('Close AI Process')
time.sleep(3)
def main():
"""MAIN"""
# Video source from webcam or video file.
video_src = args.cam if args.cam is not None else args.video
if video_src is None:
print("Warning: video source not assigned, default webcam will be used.")
video_src = 0
cap = cv2.VideoCapture(video_src)
if video_src == 0:
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = cap.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
box_process = Process(target=get_face, args=(
mark_detector, img_queue, box_queue,))
box_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
if args.out != None:
fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
output_movie = cv2.VideoWriter(args.out, fourcc, 30, (width, height))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [Stabilizer(
state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)]
tm = cv2.TickMeter()
cnt = 0
input_path = args.input_path
listdir = os.listdir(input_path)
for v_name in listdir:
v_path = os.path.join(input_path, v_name)
cap = cv2.VideoCapture(v_path)
while True:
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = cap.read()
if frame_got is False:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
if video_src == 0:
frame = cv2.flip(frame, 2)
# Pose estimation by 3 steps:
# 1. detect face;
# 2. detect landmarks;
# 3. estimate pose
# Feed frame to image queue.
img_queue.put(frame)
# Get face from box queue.
facebox = box_queue.get()
if facebox is not None:
# Detect landmarks from image of 128x128.
face_img = frame[facebox[1]: facebox[3],
facebox[0]: facebox[2]]
face_img = cv2.resize(face_img, (CNN_INPUT_SIZE, CNN_INPUT_SIZE))
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
tm.start()
marks = mark_detector.detect_marks(face_img)
tm.stop()
# Convert the marks locations from local CNN to global image.
marks *= (facebox[2] - facebox[0])
marks[:, 0] += facebox[0]
marks[:, 1] += facebox[1]
# Uncomment following line to show raw marks.
# mark_detector.draw_marks(frame, marks, color=(0, 255, 0))
# Uncomment following line to show facebox.
# mark_detector.draw_box(frame, [facebox])
# Try pose estimation with 68 points.
pose = pose_estimator.solve_pose_by_68_points(marks)
# Stabilize the pose.
steady_pose = []
pose_np = np.array(pose).flatten()
for value, ps_stb in zip(pose_np, pose_stabilizers):
ps_stb.update([value])
steady_pose.append(ps_stb.state[0])
steady_pose = np.reshape(steady_pose, (-1, 3))
# Uncomment following line to draw pose annotation on frame.
# pose_estimator.draw_annotation_box(
# frame, pose[0], pose[1], color=(255, 128, 128))
# Uncomment following line to draw stabile pose annotation on frame.
pose_estimator.draw_annotation_box(
frame, steady_pose[0], steady_pose[1], color=(128, 255, 128))
# Uncomment following line to draw head axes on frame.
# pose_estimator.draw_axes(frame, steady_pose[0], steady_pose[1])
# Show preview.
# cv2.imshow("Preview", frame)
# if cv2.waitKey(10) == 27:
# break
if args.out != None:
output_movie.write(frame)
else:
cv2.imshow("Preview", frame)
cnt = cnt + 1
if cnt % 100 == 0:
print(str(cnt), flush=True)
# Clean up the multiprocessing process.
box_process.terminate()
box_process.join()
cv2.destroyAllWindows()
if not cap.isOpened():
print('Error opening input video: {}'.format(args.video))
else:
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
# Use video writer to write processed video file.
if args.output:
video_writer = cv2.VideoWriter(
args.output, cv2.VideoWriter_fourcc('a', 'v', 'c', '1'),
cap.get(cv2.CAP_PROP_FPS),
(int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))))
# Use meter to calculate FPS.
meter = cv2.TickMeter()
# Capture frame-by-frame
while cap.isOpened():
# Start the timmer.
meter.start()
# Read a frame.
ret, frame = cap.read()
if not ret:
break
# Prepare the input image.
frame_raw = detector.preprocess(frame)
# Run the model
# tm.reset() : 시간 측정 초기화
# tm.getTimeSec() : 측정 시간을 초 단위로 반환
# tm.getTimeMilli() : 측정 시간을 밀리 초 단위로 반환
# tm.getTimeMicro() : 측정 시간을 마이크로 초 단위로 반환
import sys
import time
import numpy as np
import cv2
img = cv2.imread('ch02/hongkong.jpg')
if img is None:
print('Image load failed!')
sys.exit()
tm = cv2.TickMeter()
tm.reset()
tm.start()
t1 = time.time()
edge = cv2.Canny(img, 50, 150) # tm.start()와 tm.stop() 안에 넣어줘야 한다.
tm.stop()
ms = tm.getTimeMilli()
print('time : ', (time.time() - t1) * 1000)
print('Elapsed time : {}ms'.format(ms))
print('Resutls saved to result.jpg\n')
cv.imwrite('result.jpg', result)
# Visualize results in a new window
if args.vis:
cv.namedWindow(args.input, cv.WINDOW_AUTOSIZE)
cv.imshow(args.input, result)
cv.waitKey(0)
else: # Omit input to call default camera
deviceId = 0
cap = cv.VideoCapture(deviceId)
frameWidth = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
frameHeight = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
detector.setInputSize([frameWidth, frameHeight])
tm = cv.TickMeter()
while cv.waitKey(1) < 0:
hasFrame, frame = cap.read()
if not hasFrame:
print('No frames grabbed!')
break
# Inference
tm.start()
faces = detector.detect(frame) # faces is a tuple
tm.stop()
# Draw results on the input image
frame = visualize(frame, faces)
cv.putText(frame, 'FPS: {}'.format(tm.getFPS()), (0, 15),
import cv2
import myCV
import net
face_net = net.face_detection()
landmark_net = net.face_lanmark()
face_reid_net = net.face_reid()
body_net = myCV.Net("mo_mobilenet-ssd.xml", "mo_mobilenet-ssd.bin", (300, 300))
stream = cv2.VideoCapture(0)
counter = cv2.TickMeter()
faces_data = {}
while True:
counter.stop()
counter.start()
grab, frame = stream.read()
if not grab:
raise Exception('Image not found')
img = frame.copy()
# 15 = person id in mobilessd list
bodies = myCV.detect(body_net, frame, 0.7, 15)
for bxmin, bymin, bxmax, bymax in bodies:
cv2.rectangle(img, (bxmin, bymin), (bxmax, bymax), (255, 255, 0), 2)
bchip = frame[bymin:bymax, bxmin:bxmax]
face = myCV.detect(face_net, bchip, 0.7)
def main():
"""MAIN"""
cv2.namedWindow("Test") # Create a named window
cv2.moveWindow("Test", 900, 600) # Move it to (40,30)
screenWidth, screenHeight = pyautogui.size()
st = 'Last command'
cap = cv2.VideoCapture(0)
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
_, sample_frame = cap.read()
# Introduce mark_detector to detect landmarks.
mark_detector = MarkDetector()
# Setup process and queues for multiprocessing.
img_queue = Queue()
box_queue = Queue()
img_queue.put(sample_frame)
box_process = Process(target=get_face,
args=(
mark_detector,
img_queue,
box_queue,
))
box_process.start()
# Setting up process for listening to audio commands
voice_command_queue = Q()
stt_process = Thread(target=get_voice_command,
args=(voice_command_queue, ))
stt_process.setDaemon(True)
stt_process.start()
# Introduce pose estimator to solve pose. Get one frame to setup the
# estimator according to the image size.
height, width = sample_frame.shape[:2]
pose_estimator = PoseEstimator(img_size=(height, width))
# Introduce scalar stabilizers for pose.
pose_stabilizers = [
Stabilizer(state_num=2,
measure_num=1,
cov_process=0.1,
cov_measure=0.1) for _ in range(6)
]
tm = cv2.TickMeter()
while True:
# Read frame, crop it, flip it, suits your needs.
frame_got, frame = cap.read()
if frame_got is False:
break
# Crop it if frame is larger than expected.
# frame = frame[0:480, 300:940]
# If frame comes from webcam, flip it so it looks like a mirror.
frame = cv2.flip(frame, 2)
# Pose estimation by 3 steps:
# 1. detect face;
# 2. detect landmarks;
# 3. estimate pose
# Feed frame to image queue.
img_queue.put(frame)
# Get face from box queue.
facebox = box_queue.get()
if facebox is not None:
# Detect landmarks from image of 128x128.
face_img = frame[facebox[1]:facebox[3], facebox[0]:facebox[2]]
face_img = cv2.resize(face_img, (CNN_INPUT_SIZE, CNN_INPUT_SIZE))
face_img = cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB)
tm.start()
marks = mark_detector.detect_marks([face_img])
tm.stop()
# Convert the marks locations from local CNN to global image.
marks *= (facebox[2] - facebox[0])
marks[:, 0] += facebox[0]
marks[:, 1] += facebox[1]
# Uncomment following line to show raw marks.
# mark_detector.draw_marks(
# frame, marks, color=(0, 255, 0))
# Uncomment following line to show facebox.
# mark_detector.draw_box(frame, [facebox])
# Try pose estimation with 68 points.
pose = pose_estimator.solve_pose_by_68_points(marks)
# Stabilize the pose.
steady_pose = []
pose_np = np.array(pose).flatten()
for value, ps_stb in zip(pose_np, pose_stabilizers):
ps_stb.update([value])
steady_pose.append(ps_stb.state[0])
steady_pose = np.reshape(steady_pose, (-1, 3))
# Uncomment following line to draw pose annotation on frame.
# pose_estimator.draw_annotation_box(
# frame, pose[0], pose[1], color=(255, 128, 128))
# Uncomment following line to draw stabile pose annotation on frame.
pose_estimator.draw_annotation_box(frame,
steady_pose[0],
steady_pose[1],
color=(255, 128, 128))
# Uncomment following line to draw head axes on frame.
endpoints = pose_estimator.getEndPoints(frame, steady_pose[0],
steady_pose[1])
deltax = endpoints[1][0] - endpoints[0][0]
deltay = endpoints[1][1] - endpoints[0][1]
xpos = math.floor((deltax + 44) * screenWidth / 88)
ypos = math.floor((deltay + 14) * screenHeight / 58)
# print(xpos, ypos)
pyautogui.moveTo(xpos, ypos)
if not voice_command_queue.empty():
command = voice_command_queue.get_nowait()
if 'click' in command or 'select' in command:
pyautogui.click()
st = 'Click'
elif 'double' in command or 'in' in command:
pyautogui.doubleClick()
st = 'Double Click'
elif 'right' in command or 'menu' in command or 'light' in command:
pyautogui.rightClick()
st = 'Right Click'
print(command)
cv2.putText(frame, st, (0, 100), cv2.FONT_HERSHEY_SIMPLEX, 20, 255)
scale_percent = 30
# calculate the 50 percent of original dimensions
width = int(frame.shape[1] * scale_percent / 100)
height = int(frame.shape[0] * scale_percent / 100)
# dsize
dsize = (width, height)
# resize image
output = cv2.resize(frame, dsize)
cv2.moveWindow("Test", screenWidth - width, screenHeight - height)
# Show preview.
cv2.imshow("Test", output)
if cv2.waitKey(10) == 27:
break
# Clean up the multiprocessing process.
box_process.terminate()
box_process.join()