def __init__(self, video_file, queue_sz=128, transforms=[]):
super().__init__(queue_sz=queue_sz, transforms=transforms)
self.stream = cv.VideoCapture(get_realpath(video_file))
self.total_video_frames = int(self.stream.get(cv.CAP_PROP_FRAME_COUNT))
if self.total_video_frames <= 0:
warning(
'WARNING: Could not get the total number of frames of the video. The file may be corrupted!'
)
def __init__(self, params_file, keep_portions=True):
params = pickle.load(gzip.open(get_realpath(params_file)))
self.img_shape = params['image shape']
self.k = params['K']
self.d = params['D']
self.k_optimal = None
if keep_portions:
self.k_optimal, _ = cv.getOptimalNewCameraMatrix(
self.k, self.d, self.img_shape, 0.7, self.img_shape, True)
else:
self.k_optimal = self.k
self.map1 = None
self.map2 = None
self.imap1 = None
self.imap2 = None
self.map1, self.map2 = cv.fisheye.initUndistortRectifyMap(
self.k, self.d, None, self.k_optimal, self.img_shape, cv.CV_16SC2)
self.imap1, self.imap2 = self._get_inverse_mappings()
parser.add_argument(
'-p',
'--percent-train-samples',
type=float,
default=0.7,
help='The percentage of train samples. Default is 70%%',
)
parser.add_argument(
'--frames-ext',
type=str,
default='jpg',
help='The extension of the input frames.',
)
args = parser.parse_args()
dataset_folder = get_realpath(args.dataset_folder)
frames_folder = os.path.join(dataset_folder, 'JPEGImages')
assert os.path.isdir(frames_folder)
frames_keys = _get_frames_keys(frames_folder, args.frames_ext)
n_train = int(args.percent_train_samples * len(frames_keys))
sets_folder = os.path.join(dataset_folder, 'ImageSets', 'Main')
if not os.path.isdir(sets_folder):
os.makedirs(sets_folder)
with open(os.path.join(sets_folder, 'trainval.txt'), 'w+') as f:
for img_key in frames_keys[:n_train]:
f.write(f'{img_key}\n')
'--output-name',
type=str,
required=False,
help=
'The name of the output file containing the camera calibration parameters.',
)
parser.add_argument(
'-of',
'--output-folder',
type=str,
required=False,
help='Path to the folder where the output, file will be stored.',
)
args = parser.parse_args()
input_folder = get_realpath(args.input_folder)
if not os.path.isdir(input_folder):
print('The given input folder is not valid')
sys.exit(-1)
if args.output_folder:
output_folder = get_realpath(args.output_folder)
if not os.path.isdir(output_folder):
print('The given output folder is not valid')
sys.exit(-1)
else:
output_folder = get_realpath('./resources')
if not os.path.isdir(output_folder):
os.makedirs(output_folder)
if args.output_name:
)
parser.add_argument(
'-o',
'--output',
type=str,
required=True,
help='The output folder of the dataset.',
)
parser.add_argument(
'--frames-ext',
type=str,
default='jpg',
help='The extension of the input frames.',
)
args = parser.parse_args()
input_annos_folder = get_realpath(args.input_annos)
input_frames_folder = get_realpath(args.input_frames)
output_folder = get_realpath(args.output)
if not os.path.isdir(output_folder):
os.makedirs(output_folder)
anno_data = get_annotations_dict(
input_frames_folder, input_annos_folder, args.frames_ext
)
_save_database(
anno_data, output_folder, args.frames_ext
)
from wagon_tracking.tracking import WagonTracker
from wagon_tracking.transforms import DistortionRectifier
from wagon_tracking.utils import get_realpath
from wagon_tracking.videostream import VideoFileStream, VideoLiveStream
if len(sys.argv) < 5:
print(
'Usage: python run_ssd_example.py <net type> <model path> <label path>'
'<video file | device location> [camera_param_file]')
sys.exit(0)
net_type = sys.argv[1]
model_path = sys.argv[2]
label_path = sys.argv[3]
video_path = sys.argv[4]
if len(sys.argv) > 5:
camera_parameters = get_realpath(sys.argv[5])
transform = [DistortionRectifier(camera_parameters)]
else:
transform = []
if os.path.exists(video_path):
cap = VideoFileStream(video_path, queue_sz=64,
transforms=transform) # capture from file
else:
cap = VideoLiveStream(video_path,
transforms=transform) # capture from camera
frame_time = int(1 / cap.get(cv2.CAP_PROP_FPS) * 1000)
cap.start()
'''-------------------------- Test code --------------------------'''
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
from wagon_tracking.detection import WagonDetector
from wagon_tracking.transforms import DistortionRectifier
from wagon_tracking.videostream import VideoFileStream, VideoLiveStream
from wagon_tracking.utils import get_realpath
if len(sys.argv) < 5:
print(
'Usage: python run_ssd_example.py <net type> <model path> <label path>'
' <video file | device location> [camera_param_file]'
)
sys.exit(0)
net_type = sys.argv[1]
model_path = sys.argv[2]
label_path = sys.argv[3]
video_path = sys.argv[4]
camera_parameters = get_realpath('resources/camera_parameters.pkl.gz')
if len(sys.argv) > 5:
camera_parameters = get_realpath(sys.argv[5])
transform = [DistortionRectifier(camera_parameters)]
if os.path.exists(video_path):
cap = VideoFileStream(
video_path, queue_sz=64, transforms=transform
) # capture from file
else:
cap = VideoLiveStream(video_path, transforms=transform) # capture from camera
frame_time = int(1 / cap.get(cv2.CAP_PROP_FPS) * 1000)
cap.start()
detector = WagonDetector(net_type, label_path, model_path)