scripts/wit.py

# coding=utf-8
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# 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.
""""WIT (Wikipedia-based Image Text Dataset) dataset (Wikimedia version)."""

import base64
import gzip
import json

import datasets

from .corrected_examples import CORRECTED_EXAMPLES


_CITATION = """\
@article{srinivasan2021wit,
  title={WIT: Wikipedia-based Image Text Dataset for Multimodal Multilingual Machine Learning},
  author={Srinivasan, Krishna and Raman, Karthik and Chen, Jiecao and Bendersky, Michael and Najork, Marc},
  journal={arXiv preprint arXiv:2103.01913},
  year={2021}
}
"""

_DESCRIPTION = """\
Wikipedia-based Image Text (WIT) Dataset is a large multimodal multilingual dataset. 
It contains more than six million images from Wikipedia articles in 100+ languages, which correspond to almost all captioned images in Google's version of the WIT dataset.
Images are provided at a 300-px resolution, a size that is suitable for most of the learning frameworks used to classify and analyze images.
This version of the WIT dataset was released by Wikimedia Research team.
"""

_LICENSE = "CC BY-SA 4.0 international license"

_HOMEPAGE = "https://techblog.wikimedia.org/2021/09/09/the-wikipedia-image-caption-matching-challenge-and-a-huge-release-of-image-data-for-research/"

_BASE_URL = "https://storage.googleapis.com/huggingface-nlp/datasets/wit/"

_URLS = [_BASE_URL + f"part-{'%05d' % i}-48a6f07e-bb86-4735-aac7-883349f41a28-c000.json.gz" for i in range(400)]


class Wit(datasets.GeneratorBasedBuilder):
    """Builder for WIT dataset (Wikimedia version)."""

    DEFAULT_WRITER_BATCH_SIZE = 1000

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "image_url": datasets.Value("string"),
                    "embedding": datasets.Sequence(datasets.Value("float64"), length=2048),
                    "metadata_url": datasets.Value("string"),
                    "original_height": datasets.Value("int32"),
                    "original_width": datasets.Value("int32"),
                    "mime_type": datasets.Value("string"),
                    "caption_attribution_description": datasets.Value("string"),
                    "wit_features": datasets.Sequence(
                        {
                            "language": datasets.Value("string"),
                            "page_url": datasets.Value("string"),
                            "attribution_passes_lang_id": datasets.Value("bool"),
                            "caption_alt_text_description": datasets.Value("string"),
                            "caption_reference_description": datasets.Value("string"),
                            "caption_title_and_reference_description": datasets.Value("string"),
                            "context_page_description": datasets.Value("string"),
                            "context_section_description": datasets.Value("string"),
                            "hierarchical_section_title": datasets.Value("string"),
                            "is_main_image": datasets.Value("bool"),
                            "page_changed_recently": datasets.Value("bool"),
                            "page_title": datasets.Value("string"),
                            "section_title": datasets.Value("string"),
                        }
                    ),
                }
            ),
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        downloaded_files = dl_manager.download(_URLS)
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_files": downloaded_files}),
        ]

    def _generate_examples(self, data_files):
        """Yields examples."""
        wit_feature_names = self.info.features["wit_features"].feature.keys()
        idx = 0
        for data_file_idx, data_file in enumerate(data_files):
            with gzip.open(open(data_file, "rb"), mode="rt", encoding="utf-8") as f:
                for row_idx, row in enumerate(f):
                    example = json.loads(row)
                    ex_wit_features_non_empty = []
                    for feature in example["wit_features"]:
                        # If a feature is missing from feature dict, add it as None
                        for wit_feature_name in wit_feature_names:
                            if wit_feature_name not in feature:
                                feature[wit_feature_name] = None
                        # Here we take redundant values from wit_features and add them to example to avoid unnecessary duplication
                        extra_wit_feature_keys = [k for k in feature.keys() if k not in wit_feature_names]
                        for extra_wit_feature_key in extra_wit_feature_keys:
                            extra_wit_feature_value = feature.pop(extra_wit_feature_key)
                            if isinstance(extra_wit_feature_value, list):
                                extra_wit_feature_value = extra_wit_feature_value[0]
                            example[extra_wit_feature_key] = extra_wit_feature_value
                        # Remove empty wit features
                        if any(v is not None for v in feature.values()):
                            ex_wit_features_non_empty.append(feature)
                    example["wit_features"] = ex_wit_features_non_empty
                    # Check example now for missing keys, adding None to avoid failures
                    missing_keys = [k for k in self.info.features.keys() if k not in example]
                    for missing_key in missing_keys:
                        example[missing_key] = None
                    # Decode base64 encoded image bytes
                    b64_image_bytes = example.pop("b64_bytes")
                    example["image"] = (
                        {"path": None, "bytes": base64.b64decode(b64_image_bytes)}
                        if b64_image_bytes is not None
                        else None
                    )
                    corrections = CORRECTED_EXAMPLES.get((data_file_idx, row_idx))
                    if corrections is not None:
                        assert example["metadata_url"] == corrections["metadata_url"]
                        example.update(corrections)
                    yield idx, example
                    idx += 1