def _cache_entity_data(self, csv_file_path):
entity_set = set()
csv_data = io.csv_to_list(csv_file_path)
csv_data.pop(0) # CSV header
for csv_row in csv_data:
video_id = csv_row[0]
entity_id = csv_row[-3]
timestamp = csv_row[1]
speech_label = self._postprocess_speech_label(csv_row[-2])
entity_label = self._postprocess_entity_label(csv_row[-2])
minimal_entity_data = (entity_id, timestamp, entity_label)
# Store minimal entity data
if video_id not in self.entity_data.keys():
self.entity_data[video_id] = {}
if entity_id not in self.entity_data[video_id].keys():
self.entity_data[video_id][entity_id] = []
entity_set.add((video_id, entity_id))
self.entity_data[video_id][entity_id].append(minimal_entity_data)
#Store speech meta-data
if video_id not in self.speech_data.keys():
self.speech_data[video_id] = {}
if timestamp not in self.speech_data[video_id].keys():
self.speech_data[video_id][timestamp] = speech_label
#max operation yields if someone is speaking.
new_speech_label = max(self.speech_data[video_id][timestamp],
speech_label)
self.speech_data[video_id][timestamp] = new_speech_label
return entity_set
def _cache_entity_data_forward(self, csv_file_path, target_video):
entity_list = list()
csv_data = io.csv_to_list(csv_file_path)
csv_data.pop(0) # CSV header
for csv_row in csv_data:
video_id = csv_row[0]
if video_id != target_video:
continue
entity_id = csv_row[-3]
timestamp = csv_row[1]
entity_label = self._postprocess_entity_label(csv_row[-2])
entity_list.append((video_id, entity_id, timestamp))
minimal_entity_data = (entity_id, timestamp, entity_label
) # sfate to ingore label here
if video_id not in self.entity_data.keys():
self.entity_data[video_id] = {}
if entity_id not in self.entity_data[video_id].keys():
self.entity_data[video_id][entity_id] = []
self.entity_data[video_id][entity_id].append(minimal_entity_data)
return entity_list
def _cache_feature_file(self, csv_file):
entity_data = {}
feature_list = []
ts_to_entity = {}
print('load feature data', csv_file)
csv_data = io.csv_to_list(csv_file)
for csv_row in csv_data:
video_id = csv_row[0]
ts = csv_row[1]
entity_id = csv_row[2]
features = self._decode_feature_data_from_csv(csv_row[-1])
label = int(float(csv_row[3]))
# entity_data
if video_id not in entity_data.keys():
entity_data[video_id] = {}
if entity_id not in entity_data[video_id].keys():
entity_data[video_id][entity_id] = {}
if ts not in entity_data[video_id][entity_id].keys():
entity_data[video_id][entity_id][ts] = []
entity_data[video_id][entity_id][ts] = (features, label)
feature_list.append((video_id, entity_id, ts))
# ts_to_entity
if video_id not in ts_to_entity.keys():
ts_to_entity[video_id] = {}
if ts not in ts_to_entity[video_id].keys():
ts_to_entity[video_id][ts] = []
ts_to_entity[video_id][ts].append(entity_id)
print('loaded ', len(feature_list), ' features')
return entity_data, feature_list, ts_to_entity
ava_ground_truth_dir = '.../AVA/csv/val' #AVA original ground truth files
temporary_dir = '.../temp/activeSpeakers' #Just an empty temporary dir
# The script will generate these two, use them for the official AVA evaluation
dataset_predictions_csv = '.../Forwards/ActiveSpeakers/publish/final/STE.csv' #file with final predictions
dataset_gt_csv = '...Forwards/ActiveSpeakers/publish/final/gt.csv' # Utility file to use the official evaluation tool
#cleanup temp dir
del_files = glob.glob(temporary_dir + '/*')
for f in del_files:
os.remove(f)
pred_files, gt_files = select_files(forward_dir, ava_ground_truth_dir)
for idx, (pf, gtf) in enumerate(zip(pred_files, gt_files)):
prediction_data = csv_to_list(pf)
gt_data = csv_to_list(gtf)
print('Match', os.path.basename(pf), len(prediction_data),
len(gt_data))
if len(prediction_data) != len(gt_data):
raise Exception('Groundtruth and prediction dont match in lenght')
post_processed_predictions = prediction_postprocessing(
prediction_data, 1)
#reformat into ava required style
for idx in range(len(post_processed_predictions)):
post_processed_predictions[idx] = [
gt_data[idx][0], gt_data[idx][1], gt_data[idx][2],
gt_data[idx][3], gt_data[idx][4], gt_data[idx][5],