Spaces:
Running
on
Zero
Running
on
Zero
File size: 5,294 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 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 |
# Copyright (c) 2025 Ye Liu. Licensed under the BSD-3-Clause License.
from collections import OrderedDict
import nncore
from videomind.dataset.hybrid import DATASETS
from videomind.dataset.wrappers import GroundingDataset
from videomind.utils.parser import parse_query
@DATASETS.register(name='hirest_grounding')
class HiRESTGroundingDataset(GroundingDataset):
ANNO_PATH_TRAIN = 'data/hirest/all_data_train.json'
ANNO_PATH_VALID = 'data/hirest/all_data_val.json'
VIDEO_ROOT = 'data/hirest/videos_3fps_480_noaudio'
UNIT = 1.0
@classmethod
def load_annos(self, split='train'):
if split == 'train':
raw_annos = nncore.load(self.ANNO_PATH_TRAIN, object_pairs_hook=OrderedDict)
else:
raw_annos = nncore.load(self.ANNO_PATH_VALID, object_pairs_hook=OrderedDict)
all_videos = nncore.ls(self.VIDEO_ROOT, ext='.mp4')
all_videos = set(v[:11] for v in all_videos)
annos = []
for query, videos in raw_annos.items():
for video_name, raw_anno in videos.items():
if not raw_anno['relevant'] or not raw_anno['clip']:
continue
assert len(raw_anno['bounds']) == 2
vid = video_name.split('.')[0]
if vid not in all_videos:
continue
anno = dict(
source='hirest_grounding',
data_type='grounding',
video_path=nncore.join(self.VIDEO_ROOT, video_name),
duration=raw_anno['v_duration'],
query=parse_query(query),
span=[raw_anno['bounds']])
annos.append(anno)
return annos
@DATASETS.register(name='hirest_step')
class HiRESTStepDataset(HiRESTGroundingDataset):
@classmethod
def load_annos(self, split='train'):
if split == 'train':
raw_annos = nncore.load(self.ANNO_PATH_TRAIN, object_pairs_hook=OrderedDict)
else:
raw_annos = nncore.load(self.ANNO_PATH_VALID, object_pairs_hook=OrderedDict)
all_videos = nncore.ls(self.VIDEO_ROOT, ext='.mp4')
all_videos = set(v[:11] for v in all_videos)
annos = []
for query, videos in raw_annos.items():
for video_name, raw_anno in videos.items():
if not raw_anno['relevant'] or not raw_anno['clip'] or len(raw_anno['steps']) == 0:
continue
vid = video_name.split('.')[0]
if vid not in all_videos:
continue
for step in raw_anno['steps']:
assert len(step['absolute_bounds']) == 2
anno = dict(
source='hirest_step',
data_type='grounding',
video_path=nncore.join(self.VIDEO_ROOT, video_name),
duration=raw_anno['v_duration'],
query=parse_query(step['heading']),
span=[step['absolute_bounds']])
annos.append(anno)
return annos
@DATASETS.register(name='hirest_step_bias')
class HiRESTStepBiasDataset(HiRESTStepDataset):
@classmethod
def load_annos(self, split='train'):
if split == 'train':
raw_annos = nncore.load(self.ANNO_PATH_TRAIN, object_pairs_hook=OrderedDict)
else:
raw_annos = nncore.load(self.ANNO_PATH_VALID, object_pairs_hook=OrderedDict)
all_videos = nncore.ls(self.VIDEO_ROOT, ext='.mp4')
all_videos = set(v[:11] for v in all_videos)
annos = []
for query, videos in raw_annos.items():
for video_name, raw_anno in videos.items():
if not raw_anno['relevant'] or not raw_anno['clip'] or len(raw_anno['steps']) == 0:
continue
vid = video_name.split('.')[0]
if vid not in all_videos:
continue
for i in range(len(raw_anno['steps']) - 1):
span_a = raw_anno['steps'][i]['absolute_bounds']
span_b = raw_anno['steps'][i + 1]['absolute_bounds']
assert len(span_a) == 2 and len(span_b) == 2 and span_a[1] == span_b[0]
query_a = parse_query(f"The moment before {raw_anno['steps'][i + 1]['heading']}")
query_b = parse_query(f"The moment after {raw_anno['steps'][i]['heading']}")
anno_a = dict(
source='hirest_step_bias',
data_type='grounding',
video_path=nncore.join(self.VIDEO_ROOT, video_name),
duration=raw_anno['v_duration'],
query=query_a,
span=[span_a])
anno_b = dict(
source='hirest_step_bias',
data_type='grounding',
video_path=nncore.join(self.VIDEO_ROOT, video_name),
duration=raw_anno['v_duration'],
query=query_b,
span=[span_b])
annos.append(anno_a)
annos.append(anno_b)
return annos
|