def detect():
# dataset = ImagenetVidVRD('./vidvrd-dataset', './vidvrd-dataset/videos', ['train', 'test'])
# dataset = VidOR('./vidor-dataset/annotation', './vidor-dataset/video', ['training', 'validation'])
with open(os.path.join(get_model_path(), 'baseline_setting.json'),
'r') as fin:
param = json.load(fin)
short_term_relations = model.predict(dataset, param)
# group short term relations by video
video_st_relations = defaultdict(list)
for index, st_rel in short_term_relations.items():
vid = index[0]
video_st_relations[vid].append((index, st_rel))
# video-level visual relation detection by relational association
print('greedy relational association ...')
video_relations = dict()
for vid in tqdm(video_st_relations.keys()):
video_relations[vid] = association.greedy_relational_association(
dataset, video_st_relations[vid], max_traj_num_in_clip=100)
# save detection result
with open(
os.path.join(get_model_path(),
'baseline_relation_prediction.json'), 'w') as fout:
output = {'version': 'VERSION 1.0', 'results': video_relations}
json.dump(output, fout)
def detect():
dataset = Dataset()
with open(os.path.join(get_model_path(), 'baseline_setting.json'), 'r') as fin:
param = json.load(fin)
short_term_relations = model.predict(dataset, param)
# group short term relations by video
video_st_relations = defaultdict(list)
for index, st_rel in short_term_relations.items():
vid = index[0]
video_st_relations[vid].append((index, st_rel))
# video-level vid_features relation detection by relational association
print('greedy relational association ...')
video_relations = dict()
for vid in tqdm(video_st_relations.keys()):
video_relations[vid] = association.greedy_relational_association(
dataset, video_st_relations[vid], max_traj_num_in_clip=100)
# save detection result
with open(os.path.join(get_model_path(), 'baseline_video_relations.json'), 'w') as fout:
json.dump(video_relations, fout)