luio
July 14, 2022, 4:02pm
1
I’ve noticed that there are multiple annotations per clip, from different annotators.
Also I see no “primary” field used in the benchmark, am I missing something?
clip_annot_1[clip_id]['parent_end_sec'] = clip['video_end_sec']
clip_annot_1[clip_id]['v_duration'] = v_duration
clip_annot_1[clip_id]['fps'] = fps
clip_annot_1[clip_id]['annotations'] = []
clip_annot_1[clip_id]['subset'] = video['split']
if video['split'] != 'test':
annotations = clip['annotations']
for cnt, annot in enumerate(annotations):
for label in annot['labels']:
if label['primary']:
clip_annot_1[clip_id]['annotations'].append(label)
#######################################################################
## If there are no remaining annotations for a clip ###
## Remove the clip ###
remove_list = []
for k, v in clip_annot_1.items():
if v['subset']!='test' and len(v['annotations']) == 0:
print(f'NO annotations: video {k}')
remove_list.append(k)
Chen
July 20, 2022, 10:28am
2
Currently, we treat all annotations as ground-truth even if they are overlapped and from different annotators.
Chen
July 20, 2022, 10:30am
3
In the most recent annotations, there should be the field of primary. Please try updating your data from the server if you are not using the latest version.