File size: 1,858 Bytes
92b4fa8 | 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 | """Read frame WebDatasets using each dataset's manifest."""
import json
import io
from pathlib import Path
import numpy as np
from PIL import Image
import webdataset as wds
def decode_sample(sample):
result = dict(sample)
meta = json.loads(sample['json'])
result['json'] = meta
for modality, field in meta['components'].items():
if field is None:
continue
data = io.BytesIO(sample[field])
if field.endswith('.npy'):
result[field] = np.load(data, allow_pickle=False)
elif modality == 'depth':
# Generic RGB image decoding would destroy 16-bit depth precision.
with Image.open(data) as image:
result[field] = np.array(image)
else:
with Image.open(data) as image:
result[field] = image.convert('RGB')
return result
def dataset(directory, split='train', decode=True):
directory = Path(directory)
manifest = json.loads((directory / 'manifest.json').read_text())
if manifest['status'] != 'complete':
raise RuntimeError(f'Dataset is not complete: {manifest["status"]}')
shards = [str(directory / s['path']) for s in manifest['shards']
if s['path'].split('/')[1] == split]
if not shards:
raise ValueError(f'No shards for split {split!r}')
result = wds.WebDataset(shards, shardshuffle=False)
return result.map(decode_sample) if decode else result
if __name__ == '__main__':
import argparse
p = argparse.ArgumentParser()
p.add_argument('directory')
p.add_argument('--split', default='train')
args = p.parse_args()
sample = next(iter(dataset(args.directory, args.split)))
meta = sample['json']
print('Fields:', list(sample))
print('Dataset:', meta['dataset'], 'scene:', meta['scene_id'], 'frame:', meta['frame_id'])
|