def light_test_widerface():
# load net
cfg = widerface_640
num_classes = len(WIDERFace_CLASSES) + 1 # +1 background
net = build_ssd('test', cfg['min_dim'], num_classes) # initialize SSD
net.load_state_dict(torch.load(args.trained_model))
net.cuda()
net.eval()
print('Finished loading model!')
# load data
testset = WIDERFaceDetection(args.widerface_root, 'val', None,
WIDERFaceAnnotationTransform())
# evaluation
cuda = args.cuda
transform = TestBaseTransform((104, 117, 123))
thresh = cfg['conf_thresh']
print(thresh)
save_path = args.save_folder
num_images = len(testset)
for i in range(num_images):
img = testset.pull_image(i)
img_id, annotation = testset.pull_anno(i)
event = testset.pull_event(i)
print('Testing image {:d}/{:d} {}....'.format(i + 1, num_images,
img_id))
shrink = 1
det = infer(net, img, transform, thresh, cuda, shrink)
if not os.path.exists(save_path + event):
os.makedirs(save_path + event)
f = open(save_path + event + '/' + img_id.split(".")[0] + '.txt', 'w')
write_to_txt(f, np.array(det), event, img_id)
def light_test_oneimage():
# load net
cfg = widerface_640
num_classes = len(WIDERFace_CLASSES) + 1 # +1 background
net = build_ssd('test', cfg['min_dim'], num_classes) # initialize SSD
#net = nn.DataParallel(net)
net.load_state_dict(torch.load(args.trained_model))
net.cuda()
net.eval()
print('Finished loading model!')
# evaluation
cuda = args.cuda
transform = TestBaseTransform((104, 117, 123))
thresh = cfg['conf_thresh']
# load data
#path = './data/worlds-largest-selfie-biggest.jpg'
#path = "./data/worlds-largest-selfie.jpg"
path = "./data/yuebing.jpg"
img_id = 'result'
img = cv2.imread(path, cv2.IMREAD_COLOR)
shrink = 1
det = infer(net, img, transform, thresh, cuda, shrink)
vis_detections(img, det, img_id, 0.6)
def main():
# load net
shrink = 1
cfg = widerface_640
num_classes = len(WIDERFace_CLASSES) + 1 # +1 background
net = build_ssd('test', cfg['min_dim'], num_classes) # initialize SSD
# net = nn.DataParallel(net)
net.load_state_dict(torch.load(args.trained_model))
cuda = args.cuda
if cuda:
net.cuda()
else:
net.cpu()
net.eval()
print('Finished loading model!')
# evaluation
transform = TestBaseTransform((104, 117, 123))
thresh = cfg['conf_thresh']
video_folder_list = sorted([
filename for filename in os.listdir(args.input_folder)
if not filename.startswith('.')
]) #隠しフォルダを除く
for video in video_folder_list:
ext = os.path.splitext(os.path.split(video)[1])[1]
print(ext)
save_folder = args.input_folder.replace("input", "output")
if not os.path.exists(save_folder):
os.makedirs(save_folder)
if ext == ".mp4" or ext == ".MP4":
save_folder_tmp = save_folder + "tmp/"
save_folder_video = save_folder + "videos/"
if not os.path.exists(save_folder_tmp):
os.makedirs(save_folder_tmp)
os.makedirs(save_folder_video)
frame_id = 0
video_name = os.path.splitext(os.path.split(video)[1])[0]
# GoProなどをマスクする際には音声情報によりcv2.VideoCaptureが途中で終了することがある． よってffmpegを用いて音声を削除するコードを挟んでいる．
if args.delete_audio:
import subprocess
command_delete_audio = [
"ffmpeg", "-y", "-i", args.input_folder + video, "-vcodec",
"copy", "-map", "0:0",
args.input_folder + video.replace(".", "_an.")
]
subprocess.run(command_delete_audio)
video = video.replace(".", "_an.")
cap = cv2.VideoCapture(args.input_folder + video)
original_fps = cap.get(cv2.CAP_PROP_FPS)
while True:
_, frame = cap.read()
print(_)
if not _:
break
frame_id += 1
# if frame_id<=1358:
# continue
det = infer(net, frame, transform, thresh, cuda, shrink)
cv2.imwrite(
save_folder_tmp + video_name + '_' +
'{:09d}.jpg'.format(frame_id), frame)
print('prossing:', frame_id)
if det[0][0] == 0.1:
cv2.imwrite(
save_folder_tmp + video_name + '_' +
'{:09d}.jpg'.format(frame_id), frame)
save_mosaiced_img(frame,
det,
save_folder_tmp,
video_name,
frame_id,
scale=args.mosaic_scale,
thresh=args.visual_threshold,
kernel_size=args.mosaic_density)
save_folder_list = sorted([
filename for filename in os.listdir(save_folder_tmp)
if not filename.startswith('.')
]) #隠しフォルダを除く
print(len(save_folder_list))
print(save_folder_video)
make_video_from_images(save_folder_tmp,
save_folder_list,
os.path.join(save_folder_video,
video_name + '.mp4'),
fps=original_fps)
delete_imgs(save_folder_tmp)
elif ext == ".jpg" or ext == ".png":
save_folder_img = save_folder + "imgs/"
if not os.path.exists(save_folder_img):
os.makedirs(save_folder_img)
video_name = os.path.splitext(os.path.split(video)[1])[0]
frame_id = -1
img_path = args.input_folder + video
frame = cv2.imread(img_path)
det = infer(net, frame, transform, thresh, cuda, shrink)
if det[0][0] == 0.1:
cv2.imwrite(save_folder_img + video_name + '.jpg', frame)
save_mosaiced_img(frame,
det,
save_folder_img,
video_name,
frame_id,
scale=args.mosaic_scale,
thresh=args.visual_threshold,
kernel_size=args.mosaic_density)
delete_imgs(save_folder_tmp)
cap.get(cv2.CAP_PROP_FRAME_WIDTH) // args.reduce_scale),
int(
cap.get(cv2.CAP_PROP_FRAME_HEIGHT) //
args.reduce_scale))
out = cv2.VideoWriter(args.out, fourcc, cap.get(cv2.CAP_PROP_FPS),
out_size)
total_frame = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
f_count = 0
e_time = None
s_time = None
# load net
cfg = widerface_640
num_classes = len(WIDERFace_CLASSES) + 1 # +1 background
net = build_ssd('test', cfg['min_dim'], num_classes) # initialize SSD
# net = nn.DataParallel(net)
net.load_state_dict(torch.load(args.trained_model))
net.cuda()
net.eval()
print('Finished loading model!')
# evaluation
cuda = args.cuda
transform = TestBaseTransform((104, 117, 123))
thresh = cfg['conf_thresh']
shrink = 1
while cap.isOpened():
if args.verbose > 0 and e_time is not None: