File size: 3,790 Bytes
74e8f2f |
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 |
# Copyright 2024 Big Vision Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# pylint: disable=line-too-long
r"""Implements DocVQA in TFDS structure.
It's small data, so simple to run locally. First, copy the data to local disk.
An account will be needed in https://rrc.cvc.uab.es/?ch=17&com=downloads and
from there the task annotations and images can be fetched separatedly.
mkdir -p /tmp/data/docvqa
<COPY AND DECOMPRESS DOWNLOADED FILES HERE>
Then, run conversion locally (make sure to install tensorflow-datasets for the `tfds` util):
cd big_vision/datasets
env TFDS_DATA_DIR=/tmp/tfds tfds build --datasets=docvqa
Example to load:
import tensorflow_datasets as tfds
dataset = tfds.load('docvqa', split='val', data_dir='/tmp/tfds')
"""
import json
import os
import numpy as np
import tensorflow_datasets as tfds
_DESCRIPTION = """DocVQA dataset."""
# pylint: disable=line-too-long
_CITATION = """
@article{DBLP:journals/corr/abs-2007-00398,
author = {Minesh Mathew and
Dimosthenis Karatzas and
R. Manmatha and
C. V. Jawahar},
title = {DocVQA: {A} Dataset for {VQA} on Document Images},
journal = {CoRR},
volume = {abs/2007.00398},
year = {2020},
url = {https://arxiv.org/abs/2007.00398},
eprinttype = {arXiv},
eprint = {2007.00398},
timestamp = {Mon, 06 Jul 2020 15:26:01 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2007-00398.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
"""
# pylint: enable=line-too-long
# When running locally (recommended), copy files as above an use these:
_DOCVQA_PATH = '/tmp/data/docvqa/'
class DocVQA(tfds.core.GeneratorBasedBuilder):
"""DatasetBuilder for DocVQA dataset."""
VERSION = tfds.core.Version('1.0.0')
RELEASE_NOTES = {'1.0.0': 'First release.'}
def _info(self):
"""Returns the metadata."""
return tfds.core.DatasetInfo(
builder=self,
description=_DESCRIPTION,
features=tfds.features.FeaturesDict({
'question_id': tfds.features.Scalar(np.int32),
'image/filename': tfds.features.Text(),
'image': tfds.features.Image(encoding_format='png'),
'question': tfds.features.Text(),
'answers': tfds.features.Sequence(tfds.features.Text()),
}),
supervised_keys=None,
homepage='https://www.docvqa.org/',
citation=_CITATION,
)
def _split_generators(self, dl_manager: tfds.download.DownloadManager):
"""Returns SplitGenerators."""
return {split: self._generate_examples(split)
for split in ('val', 'train', 'test')}
def _generate_examples(self, split: str):
"""Yields (key, example) tuples from split."""
suffix = '' if split == 'test' else '_withQT'
with open(os.path.join(_DOCVQA_PATH, f'{split}_v1.0{suffix}.json')) as f:
data = json.load(f)
for v in data['data']:
question_id = v['questionId']
yield question_id, {
'question_id': question_id,
'image/filename': v['image'],
'image': os.path.join(_DOCVQA_PATH, split, v['image']),
'question': v['question'],
'answers': v.get('answers', []),
}
|