def video_demo(cam_id):
weights = args.model
detector = FaceDetector(weights)
vide_capture = cv2.VideoCapture(cam_id)
while 1:
ret, img = vide_capture.read()
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
boxes = detector(img, 0.5)
for box_index in range(boxes.shape[0]):
bbox = boxes[box_index]
color = (255, 0, 0)
thickness = 2
radius = 2
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), color, thickness)
for point_x, point_y in zip(bbox[4::2], bbox[5::2]):
cv2.circle(img_show, (int(point_x), int(point_y)), radius,
color, thickness)
cv2.namedWindow('res', 0)
cv2.imshow('res', img_show)
key = cv2.waitKey(1)
if key == ord('q'):
break
def video_demo(cam_id):
weights = args.model
detector = FaceDetector(weights)
vide_capture = cv2.VideoCapture(cam_id)
while 1:
ret, img = vide_capture.read()
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
boxes = detector(img, 0.5)
for box_index in range(boxes.shape[0]):
bbox = boxes[box_index]
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (255, 0, 0), 4)
cv2.namedWindow('res', 0)
cv2.imshow('res', img_show)
key = cv2.waitKey(1)
if key == ord('q'):
break
def demo():
args.model
detector = FaceDetector(args.model)
# files = glob.glob(data_dir + '/*')
# files.sort(key=os.path.getmtime)
files = ["/home/ubuntu/seg/TAD16K/test_3/32871.jpg"]
for pic in files:
if pic.endswith('jpg'):
img = cv2.imread(pic)
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
pred = detector(img, 0.4)
print('pred ', pred.shape, pred)
for i, pred_i in enumerate(pred):
if pred_i.shape[0] > 0:
for bbox in pred_i:
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])),
(255, 0, 255), 2)
cv2.putText(img_show, str(i),
(int(bbox[0]), int(bbox[1])),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 255),
2)
cv2.imwrite("test.png", img_show)
def image_demo(data_dir):
recognition = SignRecognition(
"/home/ubuntu/seg/autonomous-Sign-Detector/checkpoints/best_ckpt.h5")
detector = FaceDetector(args.model)
new_dir = data_dir + "_result"
files = glob.glob(data_dir + '/*')
files.sort(key=os.path.getmtime)
for pic in tqdm(files):
if pic.endswith('jpg') or pic.endswith("png"):
img = cv2.imread(pic)
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
pred = detector(img, 0.4)
for i, pred_i in enumerate(pred):
if pred_i.shape[0] > 0:
for bbox in pred_i:
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])),
(255, 0, 255), 2)
cv2.putText(img_show, str(i),
(int(bbox[0]), int(bbox[1])),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 255),
2)
newpath = pic.replace(data_dir, new_dir)
dn = os.path.dirname(newpath)
if not os.path.exists(dn):
os.makedirs(dn)
cv2.imwrite(newpath, img_show)
def image_demo(data_dir):
args.model
detector = FaceDetector(args.model)
files = glob.glob(data_dir + '/*')
files.sort(key=os.path.getmtime)
for pic in files:
if pic.endswith('jpg'):
img = cv2.imread(pic)
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
pred = detector(img, 0.4)
# print('pred ', pred.shape, pred)
for i, pred_i in enumerate(pred):
if pred_i.shape[0] > 0:
for bbox in pred_i:
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])),
(255, 0, 255), 2)
cv2.putText(img_show, str(i),
(int(bbox[0]), int(bbox[1])),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 255),
2)
cv2.namedWindow('res', 0)
cv2.imshow('res', img_show)
k = cv2.waitKey(0)
if k == 110:
continue
else:
break
def image_demo(data_dir):
args.model
detector = FaceDetector(args.model)
count = 0
pics = []
GetFileList(data_dir, pics)
pics = [x for x in pics if 'jpg' in x or 'png' in x]
#pics.sort()
for pic in pics:
img = cv2.imread(pic)
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
star = time.time()
boxes = detector(img, 0.3)
#print('one iamge cost %f s'%(time.time()-star))
#print(boxes.shape)
#print(boxes)
################toxml or json
print(boxes.shape[0])
if boxes.shape[0] == 0:
print(pic)
for box_index in range(boxes.shape[0]):
bbox = boxes[box_index]
color = (255, 0, 0)
thickness = 2
radius = 2
print(bbox)
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), color, thickness)
for point_x, point_y in zip(bbox[4::2], bbox[5::2]):
cv2.circle(img_show, (int(point_x), int(point_y)), radius,
color, thickness)
start_point = (int(bbox[4]), int(bbox[5]))
end_point = (int(bbox[6]), int(bbox[7]))
cv2.line(img_show, start_point, end_point, color, thickness)
# cv2.putText(img_show, str(bbox[4]), (int(bbox[0]), int(bbox[1]) + 30),
# cv2.FONT_HERSHEY_SIMPLEX, 1,
# (255, 0, 255), 2)
#
# cv2.putText(img_show, str(int(bbox[5])), (int(bbox[0]), int(bbox[1]) + 40),
# cv2.FONT_HERSHEY_SIMPLEX, 1,
# (0, 0, 255), 2)
cv2.namedWindow('res', 0)
cv2.imshow('res', img_show)
key = cv2.waitKey(0)
if key == ord('q'):
break
print(count)
def __init__(self):
self.face_detector = FaceDetector()
self.face_landmark = Keypoints()
self.top_k = cfg.DETECT.topk
###another thread should do detector in a slow way and update the track_box
self.track_box = None
self.previous_image = None
self.previous_box = None
self.diff_thres = 5
self.iou_thres = cfg.TRACE.iou_thres
def __init__(self):
self.face_detector = FaceDetector()
self.face_landmark = FaceLandmark()
self.trace = GroupTrack()
###another thread should run detector in a slow way and update the track_box
self.track_box = None
self.previous_image = None
self.previous_box = None
self.diff_thres = 5
self.top_k = cfg.DETECT.topk
self.iou_thres = cfg.TRACE.iou_thres
self.alpha = cfg.TRACE.smooth_box
def image_demo(data_dir):
args.model
detector = FaceDetector(args.model)
count = 0
pics = []
GetFileList(data_dir, pics)
pics = [x for x in pics if 'jpg' in x or 'png' in x]
#pics.sort()
for pic in pics:
img = cv2.imread(pic)
img_show = img.copy()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
star = time.time()
boxes = detector(img, 0.3)
#print('one iamge cost %f s'%(time.time()-star))
#print(boxes.shape)
#print(boxes)
################toxml or json
print(boxes.shape[0])
if boxes.shape[0] == 0:
print(pic)
for box_index in range(boxes.shape[0]):
bbox = boxes[box_index]
cv2.rectangle(img_show, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (255, 0, 0), 4)
# cv2.putText(img_show, str(bbox[4]), (int(bbox[0]), int(bbox[1]) + 30),
# cv2.FONT_HERSHEY_SIMPLEX, 1,
# (255, 0, 255), 2)
#
# cv2.putText(img_show, str(int(bbox[5])), (int(bbox[0]), int(bbox[1]) + 40),
# cv2.FONT_HERSHEY_SIMPLEX, 1,
# (0, 0, 255), 2)
cv2.namedWindow('res', 0)
cv2.imshow('res', img_show)
cv2.waitKey(0)
print(count)
def __init__(self):
self.face_detector = FaceDetector()
self.face_landmark = FaceLandmark()
self.trace = GroupTrack()
###another thread should run detector in a slow way and update the track_box
self.track_box = None
self.previous_image = None
self.previous_box = None
self.diff_thres = 5
self.top_k = cfg.DETECT.topk
self.min_face = cfg.DETECT.min_face
self.iou_thres = cfg.TRACE.iou_thres
self.alpha = cfg.TRACE.smooth_box
if 'ema' in cfg.TRACE.ema_or_one_euro:
self.filter = EmaFilter(self.alpha)
else:
self.filter = OneEuroFilter()
ap.add_argument( "--model", required=True, default='', help="model to eval:")
ap.add_argument( "--is_show", required=False, default=False, help="show result or not?")
ap.add_argument( "--data_dir", required=False, default="./FDDB/img", help="dir to img")
ap.add_argument( "--split_dir", required=False,default='./FDDB/FDDB-folds',help="dir to FDDB-folds")
ap.add_argument( "--result", required=False,default='./result',help="dir to write result")
args = ap.parse_args()
IMAGES_DIR = args.data_dir
ANNOTATIONS_PATH = args.split_dir
RESULT_DIR = args.result
if not os.access(RESULT_DIR,os.F_OK):
os.mkdir(RESULT_DIR)
face_detector = FaceDetector(args.model)
annotations = [s for s in os.listdir(ANNOTATIONS_PATH) if s.endswith('ellipseList.txt')]
image_lists = [s for s in os.listdir(ANNOTATIONS_PATH) if not s.endswith('ellipseList.txt')]
annotations = sorted(annotations)
image_lists = sorted(image_lists)
images_to_use = []
for n in image_lists:
with open(os.path.join(ANNOTATIONS_PATH, n)) as f:
images_to_use.extend(f.readlines())
images_to_use = [s.strip() for s in images_to_use]
with open(os.path.join(RESULT_DIR, 'faceList.txt'), 'w') as f:
for p in images_to_use:
f.write(p + '\n')
ap.add_argument("--split_dir",
required=False,
default='../FDDB/FDDB-folds',
help="dir to FDDB-folds")
ap.add_argument("--result",
required=False,
default='./resultFDDB',
help="dir to write result")
args = ap.parse_args()
IMAGES_DIR = args.data_dir
ANNOTATIONS_PATH = args.split_dir
RESULT_DIR = args.result
MODEL_PATH = args.model
face_detector = FaceDetector([MODEL_PATH])
annotations = [
s for s in os.listdir(ANNOTATIONS_PATH) if s.endswith('ellipseList.txt')
]
image_lists = [
s for s in os.listdir(ANNOTATIONS_PATH)
if not s.endswith('ellipseList.txt')
]
annotations = sorted(annotations)
image_lists = sorted(image_lists)
images_to_use = []
for n in image_lists:
with open(os.path.join(ANNOTATIONS_PATH, n)) as f:
images_to_use.extend(f.readlines())
max_score = np.max(det_accu[:, 4])
det_accu_sum = np.zeros((1, 5))
det_accu_sum[:, 0:4] = np.sum(det_accu[:, 0:4], axis=0) / np.sum(
det_accu[:, -1:])
det_accu_sum[:, 4] = max_score
try:
dets = np.row_stack((dets, det_accu_sum))
except:
dets = det_accu_sum
dets = dets[0:750, :]
return dets
#face_detector = FaceDetector([MODEL_META_PATH,MODEL_PATH])
face_detector = FaceDetector(['./model/detector.pb'])
predictions = []
for n in tqdm(images_to_use):
image_array = cv2.imread(os.path.join(IMAGES_DIR, n) + '.jpg')
image_array = cv2.cvtColor(image_array, cv2.COLOR_BGR2RGB)
# threshold is important to set low
boxes = face_detector(image_array, score_threshold=0.05)
# ##flip det but seems no needs for fddb
# flip_img=np.flip(image_array,1)
#
# boxes_flip_ = face_detector(flip_img, score_threshold=0.05)
# boxes_flip = np.zeros(boxes_flip_.shape)
# boxes_flip[:, 0] = flip_img.shape[1] - boxes_flip_[:, 2]
# boxes_flip[:, 1] = boxes_flip_[:, 1]
ap.add_argument("--split_dir",
required=False,
default='./FDDB/FDDB-folds',
help="dir to FDDB-folds")
ap.add_argument("--result",
required=False,
default='./result',
help="dir to write result")
args = ap.parse_args()
IMAGES_DIR = args.data_dir
ANNOTATIONS_PATH = args.split_dir
RESULT_DIR = args.result
MODEL_PATH = args.model
INPUT_SHAPE = (args.input_shape, args.input_shape)
face_detector = FaceDetector(MODEL_PATH)
annotations = [
s for s in os.listdir(ANNOTATIONS_PATH) if s.endswith('ellipseList.txt')
]
image_lists = [
s for s in os.listdir(ANNOTATIONS_PATH)
if not s.endswith('ellipseList.txt')
]
annotations = sorted(annotations)
image_lists = sorted(image_lists)
images_to_use = []
for n in image_lists:
with open(os.path.join(ANNOTATIONS_PATH, n)) as f:
images_to_use.extend(f.readlines())
import glob
import cv2
import os
import time
import matplotlib.pyplot as plt
import tensorflow as tf
from lib.core.api.face_detector import FaceDetector
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
detector = FaceDetector(['./model/detector_2.pb'])
def GetFileList(dir, fileList):
newDir = dir
if os.path.isfile(dir):
fileList.append(dir)
elif os.path.isdir(dir):
for s in os.listdir(dir):
# if s == "pts":
# continue
newDir = os.path.join(dir, s)
GetFileList(newDir, fileList)
return fileList
def facedetect():
count = 0
# print('t')
# data_dir = '/Users/andreyrizhiy/face_detect/ASF/*.jpg'
# pics = []