Spaces:
Running
on
Zero
Running
on
Zero
File size: 1,997 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 56 57 58 59 60 61 62 63 64 |
# Copyright (c) 2025 Ye Liu. Licensed under the BSD-3-Clause License.
import csv
import nncore
from videomind.dataset.hybrid import DATASETS
from videomind.dataset.wrappers import AnsweringDataset
from videomind.utils.parser import parse_query, parse_question
@DATASETS.register(name='nextqa')
class NExTQADataset(AnsweringDataset):
ANNO_PATH_TRAIN = 'data/nextqa/train.csv'
ANNO_PATH_VALID = 'data/nextqa/val.csv'
ANNO_PATH_TEST = 'data/nextqa/test.csv'
VIDEO_ID_MAP = 'data/nextqa/map_vid_vidorID.json'
VIDEO_ROOT = 'data/nextqa/NExTVideo'
@classmethod
def load_annos(self, split='train'):
if split == 'train':
anno_path = self.ANNO_PATH_TRAIN
elif split == 'valid':
anno_path = self.ANNO_PATH_VALID
else:
anno_path = self.ANNO_PATH_TEST
with open(anno_path, mode='r') as f:
reader = csv.DictReader(f)
raw_annos = [d for d in reader]
video_id_map = nncore.load(self.VIDEO_ID_MAP)
annos = []
for raw_anno in raw_annos:
vid = raw_anno['video']
qid = raw_anno['qid']
video_id = video_id_map[vid]
query = parse_query(raw_anno['question'].capitalize() + '?')
question = parse_question(raw_anno['question'].capitalize() + '?')
options = [raw_anno[k].capitalize() for k in ('a0', 'a1', 'a2', 'a3', 'a4')]
ans = chr(ord('A') + int(raw_anno['answer']))
answer = options[int(raw_anno['answer'])]
anno = dict(
source='nextqa',
data_type='multimodal',
uid=f'{vid}_{qid}',
video_path=nncore.join(self.VIDEO_ROOT, video_id + '.mp4'),
query=query,
question=question,
options=options,
answer=answer,
ans=ans,
task=raw_anno['type'])
annos.append(anno)
return annos
|