Spaces:
Running
on
Zero
Running
on
Zero
File size: 1,791 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 |
# Copyright (c) 2025 Ye Liu. Licensed under the BSD-3-Clause License.
import random
import nncore
import numpy as np
from videomind.dataset.hybrid import DATASETS
from videomind.dataset.wrappers import GroundingDataset
from videomind.utils.parser import parse_query
@DATASETS.register(name='didemo')
class DiDeMoDataset(GroundingDataset):
ANNO_PATH_TRAIN = 'data/didemo/train_data.json'
ANNO_PATH_VALID = 'data/didemo/val_data.json'
ANNO_PATH_TEST = 'data/didemo/test_data.json'
VIDEO_ROOT = 'data/didemo/videos_3fps_480_noaudio'
DURATIONS = 'data/didemo/durations.json'
UNIT = 1.0
@classmethod
def load_annos(self, split='train'):
if split == 'train':
raw_annos = nncore.load(self.ANNO_PATH_TRAIN)
elif split == 'valid':
raw_annos = nncore.load(self.ANNO_PATH_VALID)
else:
raw_annos = nncore.load(self.ANNO_PATH_TEST)
durations = nncore.load(self.DURATIONS)
annos = []
for raw_anno in raw_annos:
vid = raw_anno['video'].split('.')[0]
# apply mean on multiple spans
span = np.array(raw_anno['times']).mean(axis=0).tolist()
span = [round(span[0] * 5), round((span[1] + 1) * 5)]
# augment spans during training
if split == 'train':
offset = random.randint(-2, 2)
span = [span[0] + offset, span[1] + offset]
anno = dict(
source='didemo',
data_type='grounding',
video_path=nncore.join(self.VIDEO_ROOT, vid + '.mp4'),
duration=durations[vid],
query=parse_query(raw_anno['description']),
span=[span])
annos.append(anno)
return annos
|