OK-VQA.py

# Copyright 2020 The HuggingFace Datasets 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.
"""OK-VQA loading script."""


import csv
import json
import os
from pathlib import Path
import datasets


_CITATION = """\
@article{DBLP:journals/corr/abs-1906-00067,
  author    = {Kenneth Marino and
               Mohammad Rastegari and
               Ali Farhadi and
               Roozbeh Mottaghi},
  title     = {{OK-VQA:} {A} Visual Question Answering Benchmark Requiring External
               Knowledge},
  journal   = {CoRR},
  volume    = {abs/1906.00067},
  year      = {2019},
  url       = {http://arxiv.org/abs/1906.00067},
  eprinttype = {arXiv},
  eprint    = {1906.00067},
  timestamp = {Thu, 13 Jun 2019 13:36:00 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1906-00067.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}
"""


_DESCRIPTION = """\
OK-VQA is a new dataset for visual question answering that requires methods which can draw upon outside knowledge to answer questions.
- 14,055 open-ended questions
- 5 ground truth answers per question
- Manually filtered to ensure all questions require outside knowledge (e.g. from Wikipeida)
- Reduced questions with most common answers to reduce dataset bias
"""


_HOMEPAGE = "https://okvqa.allenai.org/"

# TODO: Add the licence for the dataset here if you can find it
_LICENSE = "CC BY 4.0"  # found in the zip files bellow - we show maybe ask for confirmation


_URLS = {
    "annotations": {
        "train": "https://okvqa.allenai.org/static/data/mscoco_train2014_annotations.json.zip",
        "val": "https://okvqa.allenai.org/static/data/mscoco_val2014_annotations.json.zip",
    },
    "questions": {
        "train": "https://okvqa.allenai.org/static/data/OpenEnded_mscoco_train2014_questions.json.zip",
        "val": "https://okvqa.allenai.org/static/data/OpenEnded_mscoco_val2014_questions.json.zip",
    },
    "images": {
        "train": "http://images.cocodataset.org/zips/train2014.zip",
        "val": "http://images.cocodataset.org/zips/val2014.zip",
    },
}


class OKVQADataset(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.0.0")

    def _info(self):
        features = datasets.Features(
            {
                "image": datasets.Image(),
                "question_type": datasets.Value('string'),
                "confidence": datasets.Value('int32'),
                "answers": [{
                    "answer": datasets.Value('string'),
                    "raw_answer": datasets.Value('string'),
                    "answer_confidence": datasets.Value('string'),
                    "answer_id": datasets.Value('int64'),
                }],
                "image_id": datasets.Value('int64'),
                "answer_type": datasets.Value('string'),
                "question_id": datasets.Value('int64'),
                "question": datasets.Value('string'),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        # urls = _URLS[self.config.name] # TODO later
        data_dir = dl_manager.download_and_extract(_URLS)
        gen_kwargs = {}
        for split_name in ["train", "val"]:
            gen_kwargs_per_split = {}
            for dir_name in _URLS.keys():
                if split_name in data_dir[dir_name]:
                    file_name = Path(_URLS[dir_name][split_name]).name[: -len(".zip")]
                    path = Path(data_dir[dir_name][split_name]) / file_name
                    gen_kwargs_per_split[f"{dir_name}_path"] = path
                else:
                    gen_kwargs_per_split[f"{dir_name}_path"] = None
            gen_kwargs[split_name] = gen_kwargs_per_split

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs=gen_kwargs["train"],
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs=gen_kwargs["val"],
            ),
        ]

    def _generate_examples(self, questions_path, annotations_path, images_path):
        dataset = json.load(open(annotations_path, "r"))
        questions = json.load(open(questions_path, "r"))

        qa = {ann["question_id"]: [] for ann in dataset["annotations"]}
        for ann in dataset["annotations"]:
            qa[ann["question_id"]] = ann

        for question in questions["questions"]:
            annotation = qa[question["question_id"]]
            # some checks
            assert len(set(question.keys()) ^ {"image_id", "question", "question_id"}) == 0
            assert (
                len(
                    set(annotation.keys())
                    ^ {
                            "question_type",
                            "confidence",
                            "answers",
                            "image_id",
                            "answer_type",
                            "question_id",
                        }
                )
                == 0
            )
            # build record
            record = question
            record.update(annotation)
            record["image"] = str(images_path / f"COCO_{images_path.name}_{record['image_id']:0>12}.jpg")
            yield question["question_id"], record