def eval_widerface(cfgfile, weightfile, valdir, savedir):
m = Darknet(cfgfile)
m.load_weights(weightfile)
use_cuda = 1
if use_cuda:
m.cuda()
scale_size = 16
class_names = load_class_names('data/names')
for parent,dirnames,filenames in os.walk(valdir):
if parent != valdir:
targetdir = os.path.join(savedir, os.path.basename(parent))
if not os.path.isdir(targetdir):
os.mkdir(targetdir)
for filename in filenames:
imgfile = os.path.join(parent,filename)
img = Image.open(imgfile).convert('RGB')
sized_width = int(round(img.width*1.0/scale_size) * 16)
sized_height = int(round(img.height*1.0/scale_size) * 16)
sized = img.resize((sized_width, sized_height))
print(filename, img.width, img.height, sized_width, sized_height)
if sized_width * sized_height > 1024 * 2560:
print('omit %s' % filename)
continue
boxes = do_detect(m, sized, 0.05, 0.4, use_cuda)
if True:
savename = os.path.join(targetdir, filename)
print('save to %s' % savename)
plot_boxes(img, boxes, savename, class_names)
if True:
savename = os.path.join(targetdir, os.path.splitext(filename)[0]+".txt")
print('save to %s' % savename)
save_boxes(img, boxes, savename)
def demo(cfgfile, weightfile):
model_hand = Darknet(cfgfile)
model_hand.print_network()
model_hand.load_weights(weightfile)
print('Loading weights from %s... Done!' % (weightfile))
namesfile = 'data/hands.names'
class_names = uyolo.load_class_names(namesfile)
use_cuda = 1
if use_cuda:
model_hand.cuda()
# RealSense Start
pipeline = rs.pipeline()
config = rs.config()
config.enable_stream(rs.stream.color, 640, 480, rs.format.bgr8, 30)
profile = pipeline.start(config)
# Setting exposure
s = profile.get_device().query_sensors()[1]
s.set_option(rs.option.exposure, exposure_val)
# Setting counter for evaluation
movingList = collections.deque(maxlen=100)
while True:
# Reading image from camera
frames = pipeline.wait_for_frames()
color_frame = frames.get_color_frame()
if not color_frame:
continue
img = np.asanyarray(color_frame.get_data())
if gamma_correction:
img = adjust_gamma(img, gamma=gamma_val)
# yolo stuff
sized = cv2.resize(img, (model_hand.width, model_hand.height))
bboxes = uyolo.do_detect(model_hand, sized, 0.5, 0.4, use_cuda)
print('------')
draw_img = uyolo.plot_boxes_cv2(img, bboxes, None, class_names)
# Evaluation
movingList.append(any(bboxes))
print('Continuity : {}'.format(np.mean(movingList)))
cv2.imshow(cfgfile, draw_img)
cv2.waitKey(1)
output_name = blobs.keys()[-1]
print 'output_name',output_name
return blobs[output_name]
if __name__ == '__main__':
prototxt = 'tiny_yolo_nbn_reluface.prototxt'
caffemodel = '/nfs/xiaohang/for_chenchao/tiny_yolo_nbn_reluface.caffemodel'
imgfile = 'data/face.jpg'
m = CaffeNet(prototxt, caffemodel)
use_cuda = 1
if use_cuda:
m.cuda()
img = Image.open(imgfile).convert('RGB')
sized = img.resize((m.width, m.height))
#if m.num_classes == 20:
# namesfile = '../data/voc.names'
#class_names = load_class_names(namesfile)
class_names = ['face']
for i in range(1):
start = time.time()
boxes = do_detect(m, sized, 0.5, 0.4, use_cuda)
finish = time.time()
if i == 1:
print('%s: Predicted in %f seconds.' % (imgfile, (finish-start)))
plot_boxes(img, boxes, 'predictions.jpg', class_names)
# Reading image from camera
t_start = time.time()
ret, img = cap.read()
if ret:
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
else:
continue
# Gamma(Optional) Correction
if gamma_correction:
img = adjust_gamma(img, gamma=gamma_val)
# YOLO stuff
if use_hand_tracking:
sized = cv2.resize(img, (model_hand.width, model_hand.height))
bboxes = uyolo.do_detect(model_hand, sized, hand_conf_thresh, 0.4,
use_cuda)
if any(bboxes):
center = [
int(bboxes[0][0] * test_width),
int(bboxes[0][1] * test_height)
]
img_hand_cropped, crop_box = crop_image(img, center,
hand_crop_size)
img_detection = img_hand_cropped
# print(crop_box[3])
else:
img_detection = img
else:
img_detection = img
# DOPE pose detection