File size: 1,587 Bytes
d670799 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
# Copyright (c) OpenMMLab. All rights reserved.
import json
import multiprocessing
import os
import decord
with open('HACS_v1.1.1/HACS_segments_v1.1.1.json') as f:
all_annotations = json.load(f)['database']
def parse_anno(key):
anno = {}
anno['duration_second'] = float(all_annotations[key]['duration'])
anno['annotations'] = all_annotations[key]['annotations']
anno['subset'] = all_annotations[key]['subset']
labels = set([i['label'] for i in anno['annotations']])
num_frames = int(anno['duration_second'] * 30)
for label in labels:
path = f'data/{label}/v_{key}.mp4'
if os.path.isfile(path):
vr = decord.VideoReader(path)
num_frames = len(vr)
break
anno['feature_frame'] = anno['duration_frame'] = num_frames
anno['key'] = f'v_{key}'
return anno
pool = multiprocessing.Pool(16)
video_list = list(all_annotations)
outputs = pool.map(parse_anno, video_list)
train_anno = {}
val_anno = {}
test_anno = {}
for anno in outputs:
key = anno.pop('key')
subset = anno.pop('subset')
if subset == 'training':
train_anno[key] = anno
elif subset == 'validation':
val_anno[key] = anno
else:
test_anno[key] = anno
outdir = '../../../data/HACS'
with open(f'{outdir}/hacs_anno_train.json', 'w') as f:
json.dump(train_anno, f)
with open(f'{outdir}/hacs_anno_val.json', 'w') as f:
json.dump(val_anno, f)
with open(f'{outdir}/hacs_anno_test.json', 'w') as f:
json.dump(test_anno, f)
|