clevr.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.
"""VQA v2 loading script."""


import json
from pathlib import Path
import datasets


_CITATION = """\
@inproceedings{johnson2017clevr,
  title={Clevr: A diagnostic dataset for compositional language and elementary visual reasoning},
  author={Johnson, Justin and Hariharan, Bharath and Van Der Maaten, Laurens and Fei-Fei, Li and Lawrence Zitnick, C and Girshick, Ross},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={2901--2910},
  year={2017}
}
"""

_DESCRIPTION = """\
CLEVR is a diagnostic dataset that tests a range of visual reasoning abilities. It contains minimal biases and has detailed annotations describing the kind of reasoning each question requires. We use this dataset to analyze a variety of modern visual reasoning systems, providing novel insights into their abilities and limitations.
"""

_HOMEPAGE = "https://cs.stanford.edu/people/jcjohns/clevr/"

_LICENSE = "CC BY 4.0"  # TODO need to credit both ms coco and vqa authors!

_URLS = "https://dl.fbaipublicfiles.com/clevr/CLEVR_v1.0.zip"

CLASSES = [
    "0",
    "gray",
    "cube",
    "purple",
    "yes",
    "small",
    "brown",
    "red",
    "blue",
    "7",
    "5",
    "8",
    "metal",
    "6",
    "rubber",
    "1",
    "sphere",
    "cylinder",
    "3",
    "10",
    "2",
    "yellow",
    "cyan",
    "green",
    "9",
    "large",
    "no",
    "4",
]


class ClevrDataset(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.0.0")
    DEFAULT_BUILD_CONFIG_NAME = "default"
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="default",
            version=VERSION,
            description="This config returns answers as plain text",
        ),
        datasets.BuilderConfig(
            name="classification",
            version=VERSION,
            description="This config returns answers as class labels",
        )

    ]
    def _info(self):
        if self.config.name == "classification":
            answer_feature = datasets.ClassLabel(names=CLASSES)
        else:
            answer_feature = datasets.Value("string")
        features = datasets.Features(
            {
                "question_index": datasets.Value("int64"),
                "question_family_index": datasets.Value("int64"),
                "image_filename": datasets.Value("string"),
                "split": datasets.Value("string"),
                "question": datasets.Value("string"),
                "answer": answer_feature,
                "image": datasets.Image(),
                "image_index": datasets.Value("int64"),
                "program": datasets.Sequence({
                    "inputs": datasets.Sequence(datasets.Value("int64")),
                    "function": datasets.Value("string"),
                    "value_inputs": datasets.Sequence(datasets.Value("string")),
                }),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        data_dir = dl_manager.download_and_extract(_URLS)
        gen_kwargs = {
            split_name: {
                "split": split_name,
                "questions_path": Path(data_dir) / "CLEVR_v1.0" / "questions" / f"CLEVR_{split_name}_questions.json",
                "image_folder": Path(data_dir) / "CLEVR_v1.0" / "images" / f"{split_name}",
            }
            for split_name in ["train", "val", "test"]
        }
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs=gen_kwargs["train"],
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs=gen_kwargs["val"],
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs=gen_kwargs["test"],
            ),
        ]

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

        for idx, question in enumerate(questions["questions"]):
            question["image"] = str(image_folder / f"{question['image_filename']}")
            if split == "test":
                question["question_family_index"] = -1
                question["answer"] = -1 if self.config.name == "classification" else ""
                question["program"] = [
                    {
                        "inputs": [],
                        "function": "scene",
                        "value_inputs": [],
                    }
                ]
            yield idx, question