visible-watermark-pita.py

import os
from glob import glob

import datasets
import json
from PIL import Image

_DESCRIPTION = """\
    Watermark Dataset
"""

_VERSION = datasets.Version("1.0.0")


class WatermarkPitaConfig(datasets.BuilderConfig):
    """Builder Config for Food-101"""

    def __init__(self, urls, categories, **kwargs):
        """BuilderConfig for Food-101.

        Args:
            repository: `string`, the name of the repository.
            urls: `dict<string, string>`, the urls to the data.
            categories: `list<string>`, the categories of the data.

            **kwargs: keyword arguments forwarded to super.
        """
        _VERSION = datasets.Version("1.0.0")

        super(WatermarkPitaConfig, self).__init__(version=_VERSION, **kwargs)
        self.urls = urls
        self.categories = categories


class WatermarkPita(datasets.GeneratorBasedBuilder):
    """Watermark Dataset"""

    BUILDER_CONFIGS = [
        WatermarkPitaConfig(
            name="text",
            urls={"train": "data/text/train.zip", "valid": "data/text/valid.zip"},
            categories=["text"],
        ),
        WatermarkPitaConfig(
            name="logo",
            urls={"train": "data/logo/train.zip", "valid": "data/logo/valid.zip"},
            categories=["logo"],
        ),
        WatermarkPitaConfig(
            name="mixed",
            urls={"train": "data/mixed/train.zip", "valid": "data/mixed/valid.zip"},
            categories=["logo", "text"],
        ),
    ]

    DEFAULT_CONFIG_NAME = "text"

    def _info(self):
        return datasets.DatasetInfo(
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "objects": datasets.Sequence(
                        {
                            "x": datasets.Value("float32"),
                            "y": datasets.Value("float32"),
                            "w": datasets.Value("float32"),
                            "h": datasets.Value("float32"),
                            "label": datasets.ClassLabel(names=self.config.categories),
                        }
                    ),
                }
            ),
            description=_DESCRIPTION,
        )

    def _split_generators(self, dl_manager):
        data_dir = dl_manager.download_and_extract(self.config.urls)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"split": "train", "data_dir": data_dir["train"]},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={"split": "valid", "data_dir": data_dir["valid"]},
            ),
        ]

    def _generate_examples(self, split, data_dir):
        image_dir = os.path.join(data_dir, "images")
        label_dir = os.path.join(data_dir, "labels")

        image_paths = sorted(glob(image_dir + "/*.jpg"))
        label_paths = sorted(glob(label_dir + "/*.json"))

        for idx, (image_path, label_path) in enumerate(zip(image_paths, label_paths)):
            with open(label_path, "r") as f:
                bboxes = json.load(f)

            print(bboxes)

            objects = []
            for bbox in bboxes:
                objects.append(
                    {
                        "x": bbox["x"],
                        "y": bbox["y"],
                        "w": bbox["width"],
                        "h": bbox["height"],
                        "label": bbox["label"],
                    }
                )
                objects.append(bbox)

            yield idx, {"image": image_path, "objects": objects}