AdVQA.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.
"""adVQA loading script."""


import csv
import json
import os
from pathlib import Path

import datasets


_CITATION = """\
@InProceedings{sheng2021human,
author = {Sheng, Sasha and Singh, Amanpreet and Goswami, Vedanuj and Magana, Jose Alberto Lopez and Galuba, Wojciech and Parikh, Devi and Kiela, Douwe},
title = {Human-Adversarial Visual Question Answering},
journal={arXiv preprint arXiv:2106.02280},
year = {2021},
} 
"""

_DESCRIPTION = """\
This is v1.0 of the ADVQA dataset.
"""

_HOMEPAGE = "https://adversarialvqa.org"

_LICENSE = "CC BY-NC 4.0" # In json file

_URLS = {
    "questions": {
        "val": "https://dl.fbaipublicfiles.com/advqa/v1_OpenEnded_mscoco_val2017_advqa_questions.json",
        "test-dev": "https://dl.fbaipublicfiles.com/advqa/v1_OpenEnded_mscoco_testdev2015_advqa_questions.json",
    },
    "annotations": {
        "val": "https://dl.fbaipublicfiles.com/advqa/v1_mscoco_val2017_advqa_annotations.json",
    },
    "images": {
        "val": "http://images.cocodataset.org/zips/val2014.zip",
        "test-dev": "http://images.cocodataset.org/zips/test2015.zip",
    },
}
_SUB_FOLDER_OR_FILE_NAME = {
    "questions": {
        "val": None,
        "test-dev": None,
    },
    "annotations": {
        "val": None,
    },
    "images": {
        "val": "val2014",
        "test-dev": "test2015",
    },
}


class VQAv2Dataset(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.0.0")

    # BUILDER_CONFIGS = [
    #     datasets.BuilderConfig(name="v2", version=VERSION, description="TODO later"), coco version in-domain
    #     datasets.BuilderConfig(name="v1", version=VERSION, description="TODO later"), AVQA out-of-domain 
    # ]

    def _info(self):
        features = datasets.Features(
            {
                "answers": [
                    {
                        "answer": 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"),
                "image": datasets.Image(),
            }
        )
        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 ["val", "test-dev"]:
            gen_kwargs_per_split = {} 
            for dir_name in _URLS.keys():
                sub_folder_or_file_name = _SUB_FOLDER_OR_FILE_NAME.get(dir_name, None).get(split_name, None)
                if split_name in data_dir[dir_name] and sub_folder_or_file_name is not None:
                    path = Path(data_dir[dir_name][split_name]) / sub_folder_or_file_name
                elif split_name in data_dir[dir_name]:
                    path = Path(data_dir[dir_name][split_name])
                else:
                    path = None 
                gen_kwargs_per_split[f"{dir_name}_path"] = path
            gen_kwargs[split_name] = gen_kwargs_per_split

        return [
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs=gen_kwargs["val"],
            ),
            datasets.SplitGenerator(
                name="testdev",
                gen_kwargs=gen_kwargs["test-dev"],
            ),
        ]

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

        if annotations_path is not None:
            dataset = json.load(open(annotations_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()) ^ set(["image_id", "question", "question_id"])) == 0
                assert (
                    len(
                        set(annotation.keys())
                        ^ set(
                            [
                                "answers",
                                "image_id",
                                "answer_type",
                                "question_id",
                            ]
                        )
                    )
                    == 0
                )
                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
        else:
            # No annotations for the test split
            for question in questions["questions"]:
                question.update(
                    {
                        "answers": None,
                        "answer_type": None,
                    }
                )
                question["image"] = str(images_path / f"COCO_{images_path.name}_{question['image_id']:0>12}.jpg")
                yield question["question_id"], question