|
import datasets |
|
from io import BytesIO |
|
import numpy as np |
|
|
|
_TAR_FILES=[ |
|
"data/00000.tar", |
|
"data/00001.tar", |
|
"data/00002.tar", |
|
"data/00003.tar", |
|
"data/00004.tar", |
|
"data/00005.tar", |
|
"data/00006.tar", |
|
"data/00007.tar", |
|
"data/00008.tar", |
|
"data/00009.tar", |
|
] |
|
|
|
class Food101(datasets.GeneratorBasedBuilder): |
|
"""Food-101 Images dataset.""" |
|
|
|
def _info(self): |
|
return datasets.DatasetInfo( |
|
description="TMP description", |
|
homepage="google it", |
|
citation="lmao", |
|
license="dunno, tbh, assume the worst, k thx." |
|
) |
|
|
|
def _split_generators(self, dl_manager): |
|
|
|
l=[] |
|
for k in _TAR_FILES: |
|
archive_path = dl_manager.download(k) |
|
l.append( |
|
datasets.SplitGenerator( |
|
name=k, |
|
gen_kwargs={ |
|
"npy_files": dl_manager.iter_archive(archive_path), |
|
},) |
|
) |
|
|
|
return l |
|
|
|
def _generate_examples(self, npy_files): |
|
"""Generate images and labels for splits.""" |
|
for file_path, file_obj in npy_files: |
|
|
|
numpy_bytes = file_obj.read() |
|
numpy_dict = np.load(BytesIO(numpy_bytes), allow_pickle=True) |
|
|
|
reconverted_dict = { |
|
"frames": numpy_dict.item().get("frames"), |
|
"prompt": numpy_dict.item().get("prompt") |
|
} |
|
|
|
yield file_path, { |
|
"tokenized_prompt": reconverted_dict['prompt'], |
|
"video": reconverted_dict['frames'] |
|
} |
