def add_detection( self, detection ): if not detection.timestamp in self.detections: self.detections[ detection.timestamp ] = detection # use really all candidate_ids of a detection candidates = [ x[0] for x in detection.candidate_ids ] # or use only the set of unique candidates per detection #candidates = list( set( [ x[0] for x in detection.candidate_ids ] ) ) for id in candidates: self.ids_count += 1 self.ids_sum += aux.int_id_to_binary( id ) self.ids_sum_weighted_neighbourhood += aux.weighted_neighbourhood_id( id ) self.ids_dist = self.ids_dist + aux.get_distribution_from_id( id ) # for the add_observation method we only use the first candidate, because the observations need to be independent. if self.ids_count == 0: self.ids_dists = aux.get_distribution_from_id( candidates[ 0 ] ) else: self.ids_dists = aux.add_observation( self.ids_dists, aux.get_distribution_from_id( candidates[ 0 ] ), 0.1 ) else: print 'Warning: detection not added, path already has detection for this timestamp'
def add_match(self, match):
self.matches[match.detection.timestamp] = match
if not match.detection.is_empty():
# use really all candidate_ids of a detection:
candidates = [c[0] for c in match.detection.candidate_ids]
# or use only the set of unique candidates per detection. Counterintuitively not necessarily better
#candidates = list( set( [ c[0] for c in match.detection.candidate_ids ] ) )
for c in candidates:
#self.match_ids_sum += aux.int_id_to_binary( c )
self.match_ids_sum += aux.weighted_neighbourhood_id(c)
self.match_ids_count += 1
def add_match( self, match ): self.matches[ match.detection.timestamp ] = match if not match.detection.is_empty(): # use really all candidate_ids of a detection: candidates = [ c[0] for c in match.detection.candidate_ids ] # or use only the set of unique candidates per detection. Counterintuitively not necessarily better #candidates = list( set( [ c[0] for c in match.detection.candidate_ids ] ) ) for c in candidates: #self.match_ids_sum += aux.int_id_to_binary( c ) self.match_ids_sum += aux.weighted_neighbourhood_id( c ) self.match_ids_count += 1
def get_mean_id(self):
if self.mean_id is None:
ids_sum = np.zeros(12)
ids_count = 0
for d in self.future_path:
if not d.is_empty():
candidates = [c[0] for c in d.candidate_ids]
for c in candidates:
ids_sum += aux.weighted_neighbourhood_id(c)
ids_count += 1
self.mean_id = ids_sum * 1.0 / ids_count
return self.mean_id
def get_mean_id( self ): if self.mean_id is None: ids_sum = np.zeros( 12 ) ids_count = 0 for d in self.future_path: if not d.is_empty(): candidates = [ c[0] for c in d.candidate_ids ] for c in candidates: ids_sum += aux.weighted_neighbourhood_id( c ) ids_count += 1 self.mean_id = ids_sum*1.0 / ids_count return self.mean_id
def add_detection(self, detection):
self.detections[detection.timestamp] = detection
if not detection.is_empty():
# use really all candidate_ids of a detection:
candidates = [c[0] for c in detection.candidate_ids]
# or use only the set of unique candidates per detection. Counterintuitively not necessarily better
#candidates = list( set( [ c[0] for c in detection.candidate_ids ] ) )
for c in candidates:
#self.detections_ids_sum += aux.int_id_to_binary( c )
self.detections_ids_sum += aux.weighted_neighbourhood_id(c)
self.detections_ids_count += 1
self.mean_id_needs_update = True
self.sorted_detections_need_update = True
self.sorted_unempty_detections_need_update = True
def add_detection( self, detection ): if not detection.timestamp in self.detections: self.detections[ detection.timestamp ] = detection self.ids_count += 1 self.ids_sum += detection.decoded_id self.ids_sum_mean += aux.int_id_to_binary( detection.decoded_mean ) self.ids_sum_weighted_neighbourhood += aux.weighted_neighbourhood_id( detection.decoded_mean ) self.saliency_count += detection.localizer_saliency self.ids_sum_saliency += ( np.array(detection.decoded_id) * detection.localizer_saliency ) confidence = np.min( np.abs( 0.5 - detection.decoded_id ) * 2 ) self.confidence_count += confidence self.ids_sum_confidence += ( np.array(detection.decoded_id) * confidence ) else: print 'Warning: detection not added, path already has detection for this timestamp'
def add_detection( self, detection ): self.detections[ detection.timestamp ] = detection if not detection.is_empty(): # use really all candidate_ids of a detection: candidates = [ c[0] for c in detection.candidate_ids ] # or use only the set of unique candidates per detection. Counterintuitively not necessarily better #candidates = list( set( [ c[0] for c in detection.candidate_ids ] ) ) for c in candidates: #self.detections_ids_sum += aux.int_id_to_binary( c ) self.detections_ids_sum += aux.weighted_neighbourhood_id( c ) self.detections_ids_count += 1 self.mean_id_needs_update = True self.sorted_detections_need_update = True self.sorted_unempty_detections_need_update = True
def determine_average_id_by_weighted_neighbourhood_2iter( self ): average = self.ids_sum_weighted_neighbourhood*1.0 / self.ids_count new_sum = np.zeros( 12 ) new_count = 0 for d in self.detections.values(): candidates = [ c[0] for c in d.candidate_ids ] for c in candidates: c_bin = aux.int_id_to_binary( c ) h_dis = float( np.sum( np.abs( c_bin - average ) ) ) if h_dis < 4: # here is the hamming distance limitation new_sum += aux.weighted_neighbourhood_id( c ) new_count += 1 if new_count > 0: new_average = np.round( new_sum*1.0 / new_count ) else: new_average = np.round( average ) determined_id = aux.binary_id_to_int( new_average ) return determined_id