TextCaps.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.
"""TextCaps loading script."""


import csv
import json
import os
from multiprocessing.sharedctypes import Value
from pathlib import Path

import datasets

_CITATION = """\
@article{sidorov2019textcaps,
    title={TextCaps: a Dataset for Image Captioningwith Reading Comprehension},
    author={Sidorov, Oleksii and Hu, Ronghang and Rohrbach, Marcus and Singh, Amanpreet},
    journal={arXiv preprint arXiv:2003.12462},
    year={2020}
}
"""

_DESCRIPTION = """\
extCaps requires models to read and reason about text in images to generate captions about them. Specifically, models need to incorporate a new modality of text present in the images and reason over it and visual content in the image to generate image descriptions.
Current state-of-the-art models fail to generate captions for images in TextCaps because they do not have text reading and reasoning capabilities. See the examples in the image to compare ground truth answers and corresponding predictions by a state-of-the-art model.
"""

_HOMEPAGE = "https://textvqa.org/textcaps/"

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

_URLS = {
    "captions": {
        "train": "https://dl.fbaipublicfiles.com/textvqa/data/textcaps/TextCaps_0.1_train.json",
        "val": "https://dl.fbaipublicfiles.com/textvqa/data/textcaps/TextCaps_0.1_val.json",
        "test": "https://dl.fbaipublicfiles.com/textvqa/data/textcaps/TextCaps_0.1_test.json",
    },
    "images": {
        "train": "https://dl.fbaipublicfiles.com/textvqa/images/train_val_images.zip",
        "val": "https://dl.fbaipublicfiles.com/textvqa/images/train_val_images.zip",
        "test": "https://dl.fbaipublicfiles.com/textvqa/images/test_images.zip",
    },
    "ocr_tokens": {
        "train": "https://dl.fbaipublicfiles.com/textvqa/data/TextVQA_Rosetta_OCR_v0.2_train.json",
        "val": "https://dl.fbaipublicfiles.com/textvqa/data/TextVQA_Rosetta_OCR_v0.2_val.json",
        "test": "https://dl.fbaipublicfiles.com/textvqa/data/TextVQA_Rosetta_OCR_v0.2_test.json",
    },
}

_SUB_FOLDER_OR_FILE_NAME = {
    "images": {
        "train": "train_images",
        "val": "train_images",
        "test": "test_images",
    },
}


class TextCapsDataset(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("1.0.0")

    # BUILDER_CONFIGS = [
    #     datasets.BuilderConfig(name="v2", version=VERSION, description="TODO later"),
    #     datasets.BuilderConfig(name="v1", version=VERSION, description="TODO later"),
    # ]

    def _info(self):
        features = datasets.Features(
            {
                "ocr_tokens": [datasets.Value("string")],
                "ocr_info": [
                    {
                        "word": datasets.Value("string"),
                        "bounding_box": {
                            "width": datasets.Value("float"),
                            "height": datasets.Value("float"),
                            "rotation": datasets.Value("float"),
                            "roll": datasets.Value("float"),
                            "pitch": datasets.Value("float"),
                            "yaw": datasets.Value("float"),
                            "top_left_x": datasets.Value("float"),
                            "top_left_y": datasets.Value("float"),
                        },
                    }
                ],
                "image": datasets.Image(),
                "image_id": datasets.Value("string"),
                "image_classes": [datasets.Value("string")],
                "flickr_original_url": datasets.Value("string"),
                "flickr_300k_url": datasets.Value("string"),
                "image_width": datasets.Value("int32"),
                "image_height": datasets.Value("int32"),
                "set_name": datasets.Value("string"),
                "image_name": datasets.Value("string"),
                "image_path": datasets.Value("string"),
                "reference_strs": [datasets.Value("string")],
                "reference_tokens": [[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 = {
            split_name: {
                f"{dir_name}_path": Path(data_dir[dir_name][split_name])
                if split_name in data_dir[dir_name]
                else None
                for dir_name in _URLS.keys()
            }
            for split_name in ["train", "val", "test"]
        }

        for split_name in ["train", "val", "test"]:
            gen_kwargs[split_name]["split_name"] = split_name
            gen_kwargs[split_name]["images_path"] = (
                gen_kwargs[split_name]["images_path"]
                / _SUB_FOLDER_OR_FILE_NAME["images"][split_name]
            )
        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, captions_path, ocr_tokens_path, images_path, split_name
    ):
        seen_image_ids = set()
        captions = json.load(open(captions_path, "r"))["data"]
        ocr_tokens = json.load(open(ocr_tokens_path, "r"))["data"]

        ocr_tokens_per_image_id = {}
        for ocr_item in ocr_tokens:
            ocr_tokens_per_image_id[ocr_item["image_id"]] = ocr_item

        for caption_item in captions:
            if caption_item["image_id"] in seen_image_ids:
                continue
            seen_image_ids.add(caption_item["image_id"])
            ocr_item = ocr_tokens_per_image_id[caption_item["image_id"]]
            record = {
                "ocr_tokens": ocr_item["ocr_tokens"],
                "ocr_info": ocr_item["ocr_info"],
                "image_id": caption_item["image_id"],
                "image_classes": caption_item["image_classes"],
                "flickr_original_url": caption_item["flickr_original_url"],
                "flickr_300k_url": caption_item["flickr_300k_url"],
                "image_width": caption_item["image_width"],
                "image_height": caption_item["image_height"],
                "set_name": caption_item["set_name"],
                "image_name": caption_item["image_name"],
                "image_path": caption_item["image_path"],
                "image" : str(images_path / f'{caption_item["image_name"]}.jpg')
            }
            if not split_name == "test":
                record["reference_strs"] = caption_item["reference_strs"]
                record["reference_tokens"] = caption_item["reference_tokens"]
            else:
                record["reference_strs"] = None
                record["reference_tokens"] = None
            yield caption_item["image_id"], record