def __init__(self,
gt_path,
video_path,
metric_func,
init=True,
classifier=None):
self.classifier = classifier
self.capture = create_capture(video_path)
self.converter = VideoFrameConverter(self.capture)
self.ground_truth = GTHVideoSubstrates(gt_path, self.capture)
self.classes = self.ground_truth.classes_present
self.metric = metric_func
self.training_frames = [] # Stores all frames used in training
self.x = [] # frames n_samples * n_features
self.y = [] # labels n_samples
self.classifier.ground_truth = gt_path
self.classifier.video_path = video_path
self.training_c_to_frames = {}
self._setup_training_dict()
if init:
self.train()
def load_video(self):
path = self.lbl_path.text()
self.cap = create_capture(path)
self.vid = VideoFrameConverter(self.cap)
self.sld_pos.setRange(0, self.vid.num_frames)
self.sld_pos.setValue(0)
self.timer = QTimer(self)
self.timer.timeout.connect(self._play)
self.timer.start(27)
self.update()
def __init__(self,
capture,
video_src,
output_dir,
annotator,
existing_data=None):
"""
frameskip at 700 is *roughly* 30seconds
"""
self._capture = capture
self._converter = VideoFrameConverter(capture)
self._frames_num_total = self._converter.num_frames
self._output_dir = output_dir
self._video_src = video_src
self._complete = False
self._existing_data = existing_data
self.set_annotator(annotator)
self._ground_truth = {
STR_AUTH: annotator,
STR_DATE: time.time(),
STR_VERS: APP_VERSION,
STR_COMP: False,
STR_FILE: video_src,
STR_DATA: {},
STR_EXST: None
}
if self._existing_data is not None:
self._build_from_existing_data()
class ValidWindow(unittest.TestCase):
def setUp(self):
path_gp = p.get_path(p.KEY_GOPRO)
self.fp = path_gp + "Site 2.MP4"
self.cap = create_capture(self.fp)
self.conv = VideoFrameConverter(self.cap)
def test_site_2(self):
patches = self.conv.get_frame_mxn_patches_generator(20, 20)
assert True
def load_video(self):
path = self.lbl_path.text()
self.cap = create_capture(path)
self.vid = VideoFrameConverter(self.cap)
self.sld_pos.setRange(0, self.vid.num_frames)
self.sld_pos.setValue(0)
self.timer = QTimer(self)
self.timer.timeout.connect(self._play)
self.timer.start(27)
self.update()
class ValidWindow(unittest.TestCase):
def setUp(self):
path_gp = p.get_path(p.KEY_GOPRO)
self.fp = path_gp + "Site 2.MP4"
self.cap = create_capture(self.fp)
self.conv = VideoFrameConverter(self.cap)
def test_site_2(self):
patches = self.conv.get_frame_mxn_patches_generator(20, 20)
assert True
def __init__(self, gt_path, video_path,
metric_func,
init=True, classifier=None):
self.classifier = classifier
self.capture = create_capture(video_path)
self.converter = VideoFrameConverter(self.capture)
self.ground_truth = GTHVideoSubstrates(gt_path, self.capture)
self.classes = self.ground_truth.classes_present
self.metric = metric_func
self.training_frames = [] # Stores all frames used in training
self.x = [] # frames n_samples * n_features
self.y = [] # labels n_samples
self.classifier.ground_truth = gt_path
self.classifier.video_path = video_path
self.training_c_to_frames = {}
self._setup_training_dict()
if init:
self.train()
class EXPMLVidSubstrateBase(object):
def __init__(self, gt_path, video_path,
metric_func,
init=True, classifier=None):
self.classifier = classifier
self.capture = create_capture(video_path)
self.converter = VideoFrameConverter(self.capture)
self.ground_truth = GTHVideoSubstrates(gt_path, self.capture)
self.classes = self.ground_truth.classes_present
self.metric = metric_func
self.training_frames = [] # Stores all frames used in training
self.x = [] # frames n_samples * n_features
self.y = [] # labels n_samples
self.classifier.ground_truth = gt_path
self.classifier.video_path = video_path
self.training_c_to_frames = {}
self._setup_training_dict()
if init:
self.train()
def _setup_training_dict(self):
for k in self.classes:
if k not in self.training_c_to_frames:
self.training_c_to_frames[k] = [x for x in self.get_training_frames(k)]
# def process_frame(self, frame):
# """
# Default frame processor, simply uses self.metric function pointer on
# the entire frame passed through. Converts the representation to a
# n x 1 vector for speed by default. Returns processed data in the form
# of a generator.
# """
# data = self.metric(frame)
#
# #if isinstance(data, np.ndarray):
# # data = data.ravel()
#
# yield data
def train(self):
"""
Trains the classifier, looks to process_frame to deliver the metric-affected
version of the whole frame, or subsections thereof. This method doesn't
really care if process_frame returns 1 or more results, it just collates
the training vectors X and Y before passing to sklearn for fitting.
"""
# Thread safe lists here? @todo: multiprocessing
for k, v in self.training_c_to_frames.iteritems():
for i in v:
self.converter.set_current_frame(i)
self.x.append(self.metric(self.converter.capture_next_frame(True)))
self.y.append(k)
log.info("Beginning fitting for " + self.classifier.identifier)
# Keep a reference to the frames we have chosen for this iteration
self.classifier.training_data = self.training_frames
self.classifier.raw_classifier.fit(np.array(self.x), np.array(self.y))
def process_frame(self, frame):
return self.metric(frame)
def get_training_frames(self, klass, n=None):
"""
Generator method that records the training frames selected in this iteration,
and yields the frames sequentially for given klass. Parameter n is optional,
and may indeed be deprecated. It is to limit the number below 10% of all frames,
and was only relevant for Gabor full-frame image representations.
"""
frames, _ = self.ground_truth.get_sample_frames_for_class(klass, n=n)
self.training_frames.extend(frames) # Keep a reference to them
for f in frames:
yield f
def get_data_from_frame(self, frame):
return frame
class WTrackerMarking(QMainWindow):
"""
Main Tracking Window
"""
STR_WIN_TITLE = 'Mark Video'
STR_CLK = 'clicked()'
STR_SPECIFY_VIDEO = 'Specify Video'
def __init__(self, def_path=None):
super(WTrackerMarking, self).__init__(None)
self._trackers = HLDTrackers()
self.paused = False
self.timer = QTimer(self)
# Widgets
lbl_path_str = QLabel("Video Path:")
path = def_path if def_path is not None else os.getcwd()
self.lbl_path = QLabel(path)
self.lbl_path.setFrameStyle(QFrame.StyledPanel | QFrame.Sunken)
lbl_sld_pos = QLabel("Position:")
btn_path = QPushButton("&Path...")
btn_load = QPushButton("&Load")
btn_vid_toggle = QPushButton("&Start/Stop")
btn_reset_video = QPushButton("&Reset Video")
btn_reset_track = QPushButton("Reset &Trackers")
self.sld_pos = QWVideoPositionSlider(Qt.Horizontal, self, self.pause,
self.unpause)
self.sld_pos.valueChanged[int].connect(self.update_position)
btn_rem_tracker = QPushButton("&Delete")
self.cmb_trackers = QComboBox(self)
self.cmb_tracker_types = QComboBox(self)
self._build_cmb_trackers()
# Images
self.win_vid_filters = WTrackerOutputWindow(self)
self.frm_vid_output = QWVideoFrameLabel(self)
self.frm_vid_output.setGeometry(10, 10, 400, 100)
# Callbacks
self.connect(btn_path, SIGNAL(self.STR_CLK), self.set_path)
self.connect(btn_load, SIGNAL(self.STR_CLK), self.load_video)
self.connect(btn_vid_toggle, SIGNAL(self.STR_CLK), self.toggle_video)
self.connect(btn_reset_video, SIGNAL(self.STR_CLK),
self.do_reset_video)
self.connect(btn_reset_track, SIGNAL(self.STR_CLK),
self.do_reset_trackers)
self.connect(btn_rem_tracker, SIGNAL(self.STR_CLK),
self.remove_tracker)
# Layouts
lyt_main = QVBoxLayout()
lyt_top = QHBoxLayout()
lyt_top.addWidget(lbl_path_str)
lyt_top.addWidget(self.lbl_path, 1)
lyt_top.addWidget(btn_path)
lyt_btm = QHBoxLayout()
lyt_btm.addWidget(btn_load)
lyt_btm.addWidget(btn_vid_toggle)
lyt_btm.addWidget(btn_reset_video)
lyt_btm.addWidget(btn_reset_track)
lyt_btm.addWidget(self.cmb_tracker_types)
lyt_btm.addWidget(lbl_sld_pos)
lyt_btm.addWidget(self.sld_pos)
lyt_btm.addWidget(self.cmb_trackers)
lyt_btm.addWidget(btn_rem_tracker)
lyt_main.setAlignment(Qt.AlignTop)
lyt_main.addLayout(lyt_top)
lyt_main.addLayout(lyt_btm)
lyt_main.addWidget(self.frm_vid_output)
lyt_main.setMargin(10)
lyt_main.setSpacing(10)
widget = QWidget()
widget.setLayout(lyt_main)
self.setCentralWidget(widget)
self.setGeometry(260, 20, 1000, 55)
self.setWindowTitle(self.STR_WIN_TITLE)
## Overridden functions, handle child window
def show(self):
super(WTrackerMarking, self).show()
self.win_vid_filters.show()
def hide(self):
self.win_vid_filters.hide()
super(WTrackerMarking, self).hide()
def close(self):
self.win_vid_filters.close()
super(WTrackerMarking, self).close()
## Object functions
def _build_cmb_trackers(self, default='MOSSE'):
for t in TRACKER_TYPES.keys():
self.cmb_tracker_types.addItem(t)
id_ref = self.cmb_tracker_types.findText(default)
self.cmb_tracker_types.setCurrentIndex(id_ref)
def add_tracker(self, rect):
tracker_name = self.cmb_tracker_types.currentText()
tracker = TRACKER_TYPES[str(tracker_name)]
frame_gray = cvtColor(self.vid.current_frame, COLOR_BGR2GRAY)
# This now depends on the value of the combobox
new_tracker = tracker(frame_gray, rect)
self.trackers.add(new_tracker)
tracker_id = str(new_tracker.get_id())
self.cmb_trackers.addItem(tracker_id)
id_ref = self.cmb_trackers.findText(tracker_id)
self.cmb_trackers.setCurrentIndex(id_ref)
def remove_tracker(self):
tracker_id = self.current_tracker_id
self.trackers.remove(tracker_id)
tracker_ref = self.cmb_trackers.findText(tracker_id)
self.cmb_trackers.removeItem(tracker_ref)
def do_reset_video(self):
self.timer.stop()
self.load_video()
def do_reset_trackers(self):
self.trackers.reset()
self.cmb_trackers.clear()
def toggle_video(self):
self.paused = not self.paused
def set_path(self):
path = QFileDialog.getOpenFileName(self, self.STR_SPECIFY_VIDEO,
self.lbl_path.text())
if path:
self.lbl_path.setText(QDir.toNativeSeparators(path))
def load_video(self):
path = self.lbl_path.text()
self.cap = create_capture(path)
self.vid = VideoFrameConverter(self.cap)
self.sld_pos.setRange(0, self.vid.num_frames)
self.sld_pos.setValue(0)
self.timer = QTimer(self)
self.timer.timeout.connect(self._play)
self.timer.start(27)
self.update()
def pause(self):
self.paused = True
def unpause(self):
self.paused = False
def _play(self):
if self.paused:
return
frame = self.vid.get_next_frame()
if frame:
if self.trackers.active:
raw_frame = self.vid.current_frame
frame_gray = cvtColor(raw_frame, COLOR_BGR2GRAY)
self.trackers.update_all(frame_gray)
self.trackers.write_to_frame(raw_frame)
frame = convert_single_frame(raw_frame)
tracker = self.trackers.get(self.current_tracker_id)
vis = cvtColor(tracker.state_vis, COLOR_GRAY2RGB)
q_vis = convert_single_frame(vis)
self.win_vid_filters.update_image(q_vis)
self.frm_vid_output.setPixmap(frame)
else:
self.timer.stop()
self.load_video()
def update_position(self, value):
self.vid.set_current_frame(value)
@property
def trackers(self):
return self._trackers
@trackers.setter
def trackers(self, new_trackers):
self._trackers = new_trackers
@property
def current_tracker_id(self):
return str(self.cmb_trackers.currentText())
def __init__(self, input_json, capture):
self._input_json = input_json
self._capture = capture
self._converter = VideoFrameConverter(capture)
self._classes = {}
self._parse_json(input_json)
class WTrackerMarking(QMainWindow):
"""
Main Tracking Window
"""
STR_WIN_TITLE = 'Mark Video'
STR_CLK = 'clicked()'
STR_SPECIFY_VIDEO = 'Specify Video'
def __init__(self, def_path=None):
super(WTrackerMarking, self).__init__(None)
self._trackers = HLDTrackers()
self.paused = False
self.timer = QTimer(self)
# Widgets
lbl_path_str = QLabel("Video Path:")
path = def_path if def_path is not None else os.getcwd()
self.lbl_path = QLabel(path)
self.lbl_path.setFrameStyle(QFrame.StyledPanel | QFrame.Sunken)
lbl_sld_pos = QLabel("Position:")
btn_path = QPushButton("&Path...")
btn_load = QPushButton("&Load")
btn_vid_toggle = QPushButton("&Start/Stop")
btn_reset_video = QPushButton("&Reset Video")
btn_reset_track = QPushButton("Reset &Trackers")
self.sld_pos = QWVideoPositionSlider(Qt.Horizontal, self, self.pause, self.unpause)
self.sld_pos.valueChanged[int].connect(self.update_position)
btn_rem_tracker = QPushButton("&Delete")
self.cmb_trackers = QComboBox(self)
self.cmb_tracker_types = QComboBox(self)
self._build_cmb_trackers()
# Images
self.win_vid_filters = WTrackerOutputWindow(self)
self.frm_vid_output = QWVideoFrameLabel(self)
self.frm_vid_output.setGeometry(10, 10, 400, 100)
# Callbacks
self.connect(btn_path, SIGNAL(self.STR_CLK), self.set_path)
self.connect(btn_load, SIGNAL(self.STR_CLK), self.load_video)
self.connect(btn_vid_toggle, SIGNAL(self.STR_CLK), self.toggle_video)
self.connect(btn_reset_video, SIGNAL(self.STR_CLK), self.do_reset_video)
self.connect(btn_reset_track, SIGNAL(self.STR_CLK), self.do_reset_trackers)
self.connect(btn_rem_tracker, SIGNAL(self.STR_CLK), self.remove_tracker)
# Layouts
lyt_main = QVBoxLayout()
lyt_top = QHBoxLayout()
lyt_top.addWidget(lbl_path_str)
lyt_top.addWidget(self.lbl_path, 1)
lyt_top.addWidget(btn_path)
lyt_btm = QHBoxLayout()
lyt_btm.addWidget(btn_load)
lyt_btm.addWidget(btn_vid_toggle)
lyt_btm.addWidget(btn_reset_video)
lyt_btm.addWidget(btn_reset_track)
lyt_btm.addWidget(self.cmb_tracker_types)
lyt_btm.addWidget(lbl_sld_pos)
lyt_btm.addWidget(self.sld_pos)
lyt_btm.addWidget(self.cmb_trackers)
lyt_btm.addWidget(btn_rem_tracker)
lyt_main.setAlignment(Qt.AlignTop)
lyt_main.addLayout(lyt_top)
lyt_main.addLayout(lyt_btm)
lyt_main.addWidget(self.frm_vid_output)
lyt_main.setMargin(10)
lyt_main.setSpacing(10)
widget = QWidget()
widget.setLayout(lyt_main)
self.setCentralWidget(widget)
self.setGeometry(260, 20, 1000, 55)
self.setWindowTitle(self.STR_WIN_TITLE)
## Overridden functions, handle child window
def show(self):
super(WTrackerMarking, self).show()
self.win_vid_filters.show()
def hide(self):
self.win_vid_filters.hide()
super(WTrackerMarking, self).hide()
def close(self):
self.win_vid_filters.close()
super(WTrackerMarking, self).close()
## Object functions
def _build_cmb_trackers(self, default='MOSSE'):
for t in TRACKER_TYPES.keys():
self.cmb_tracker_types.addItem(t)
id_ref = self.cmb_tracker_types.findText(default)
self.cmb_tracker_types.setCurrentIndex(id_ref)
def add_tracker(self, rect):
tracker_name = self.cmb_tracker_types.currentText()
tracker = TRACKER_TYPES[str(tracker_name)]
frame_gray = cvtColor(self.vid.current_frame, COLOR_BGR2GRAY)
# This now depends on the value of the combobox
new_tracker = tracker(frame_gray, rect)
self.trackers.add(new_tracker)
tracker_id = str(new_tracker.get_id())
self.cmb_trackers.addItem(tracker_id)
id_ref = self.cmb_trackers.findText(tracker_id)
self.cmb_trackers.setCurrentIndex(id_ref)
def remove_tracker(self):
tracker_id = self.current_tracker_id
self.trackers.remove(tracker_id)
tracker_ref = self.cmb_trackers.findText(tracker_id)
self.cmb_trackers.removeItem(tracker_ref)
def do_reset_video(self):
self.timer.stop()
self.load_video()
def do_reset_trackers(self):
self.trackers.reset()
self.cmb_trackers.clear()
def toggle_video(self):
self.paused = not self.paused
def set_path(self):
path = QFileDialog.getOpenFileName(self,
self.STR_SPECIFY_VIDEO, self.lbl_path.text())
if path:
self.lbl_path.setText(QDir.toNativeSeparators(path))
def load_video(self):
path = self.lbl_path.text()
self.cap = create_capture(path)
self.vid = VideoFrameConverter(self.cap)
self.sld_pos.setRange(0, self.vid.num_frames)
self.sld_pos.setValue(0)
self.timer = QTimer(self)
self.timer.timeout.connect(self._play)
self.timer.start(27)
self.update()
def pause(self):
self.paused = True
def unpause(self):
self.paused = False
def _play(self):
if self.paused:
return
frame = self.vid.get_next_frame()
if frame:
if self.trackers.active:
raw_frame = self.vid.current_frame
frame_gray = cvtColor(raw_frame, COLOR_BGR2GRAY)
self.trackers.update_all(frame_gray)
self.trackers.write_to_frame(raw_frame)
frame = convert_single_frame(raw_frame)
tracker = self.trackers.get(self.current_tracker_id)
vis = cvtColor(tracker.state_vis, COLOR_GRAY2RGB)
q_vis = convert_single_frame(vis)
self.win_vid_filters.update_image(q_vis)
self.frm_vid_output.setPixmap(frame)
else:
self.timer.stop()
self.load_video()
def update_position(self, value):
self.vid.set_current_frame(value)
@property
def trackers(self):
return self._trackers
@trackers.setter
def trackers(self, new_trackers):
self._trackers = new_trackers
@property
def current_tracker_id(self):
return str(self.cmb_trackers.currentText())
def setUp(self):
path_gp = p.get_path(p.KEY_GOPRO)
self.fp = path_gp + "Site 2.MP4"
self.cap = create_capture(self.fp)
self.conv = VideoFrameConverter(self.cap)
class EXPMLVidSubstrateBase(object):
def __init__(self,
gt_path,
video_path,
metric_func,
init=True,
classifier=None):
self.classifier = classifier
self.capture = create_capture(video_path)
self.converter = VideoFrameConverter(self.capture)
self.ground_truth = GTHVideoSubstrates(gt_path, self.capture)
self.classes = self.ground_truth.classes_present
self.metric = metric_func
self.training_frames = [] # Stores all frames used in training
self.x = [] # frames n_samples * n_features
self.y = [] # labels n_samples
self.classifier.ground_truth = gt_path
self.classifier.video_path = video_path
self.training_c_to_frames = {}
self._setup_training_dict()
if init:
self.train()
def _setup_training_dict(self):
for k in self.classes:
if k not in self.training_c_to_frames:
self.training_c_to_frames[k] = [
x for x in self.get_training_frames(k)
]
# def process_frame(self, frame):
# """
# Default frame processor, simply uses self.metric function pointer on
# the entire frame passed through. Converts the representation to a
# n x 1 vector for speed by default. Returns processed data in the form
# of a generator.
# """
# data = self.metric(frame)
#
# #if isinstance(data, np.ndarray):
# # data = data.ravel()
#
# yield data
def train(self):
"""
Trains the classifier, looks to process_frame to deliver the metric-affected
version of the whole frame, or subsections thereof. This method doesn't
really care if process_frame returns 1 or more results, it just collates
the training vectors X and Y before passing to sklearn for fitting.
"""
# Thread safe lists here? @todo: multiprocessing
for k, v in self.training_c_to_frames.iteritems():
for i in v:
self.converter.set_current_frame(i)
self.x.append(
self.metric(self.converter.capture_next_frame(True)))
self.y.append(k)
log.info("Beginning fitting for " + self.classifier.identifier)
# Keep a reference to the frames we have chosen for this iteration
self.classifier.training_data = self.training_frames
self.classifier.raw_classifier.fit(np.array(self.x), np.array(self.y))
def process_frame(self, frame):
return self.metric(frame)
def get_training_frames(self, klass, n=None):
"""
Generator method that records the training frames selected in this iteration,
and yields the frames sequentially for given klass. Parameter n is optional,
and may indeed be deprecated. It is to limit the number below 10% of all frames,
and was only relevant for Gabor full-frame image representations.
"""
frames, _ = self.ground_truth.get_sample_frames_for_class(klass, n=n)
self.training_frames.extend(frames) # Keep a reference to them
for f in frames:
yield f
def get_data_from_frame(self, frame):
return frame
def setUp(self):
path_gp = p.get_path(p.KEY_GOPRO)
self.fp = path_gp + "Site 2.MP4"
self.cap = create_capture(self.fp)
self.conv = VideoFrameConverter(self.cap)