Spaces:
Running
on
Zero
Running
on
Zero
File size: 1,703 Bytes
6073e55 23fdbc0 |
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 |
# Copyright (c) 2025 Ye Liu. Licensed under the BSD-3-Clause License.
import nncore
from videomind.dataset.hybrid import DATASETS
from videomind.dataset.wrappers import AnsweringCropDataset
from videomind.utils.parser import parse_query, parse_question
@DATASETS.register(name='star')
class STARDataset(AnsweringCropDataset):
ANNO_PATH_TRAIN = 'data/star/STAR_train.json'
ANNO_PATH_VALID = 'data/star/STAR_val.json'
VIDEO_ROOT = 'data/charades_sta/videos_3fps_480_noaudio'
DURATIONS = 'data/charades_sta/durations.json'
UNIT = 0.1
@classmethod
def load_annos(self, split='train'):
if split == 'train':
raw_annos = nncore.load(self.ANNO_PATH_TRAIN)
else:
raw_annos = nncore.load(self.ANNO_PATH_VALID)
durations = nncore.load(self.DURATIONS)
annos = []
for raw_anno in raw_annos:
vid = raw_anno['video_id']
options = [c['choice'] for c in raw_anno['choices']]
answer = raw_anno['answer']
ans = chr(ord('A') + options.index(answer))
anno = dict(
source='star',
data_type='multimodal',
video_path=nncore.join(self.VIDEO_ROOT, vid + '.mp4'),
duration=durations[vid],
query=parse_query(raw_anno['question']),
question=parse_question(raw_anno['question']),
options=options,
answer=answer,
ans=ans,
span=[[raw_anno['start'], raw_anno['end']]],
task=raw_anno['question_id'].split('_')[0],
no_aug=True)
annos.append(anno)
return annos
|