pcbm_metashift.py

import datasets
import json
import itertools
from string import Template
from pathlib import Path

_HOMEPAGE = ""
_CITATION = ""
_LICENSE = ""
_DESCRIPTION_TEMPLATE = Template(
    "$num_classes-way image classification task "
    "to test domain shift of class $spurious_class from "
    "context $source_context to $target_context. "
    "Selected classes: $selected_classes"
)
_REPO = "https://huggingface.co/datasets/fact-40/pcbm_metashift/resolve/main"
_IMAGES_DIR = Path("data")


class PCBMMetashiftConfig(datasets.BuilderConfig):
    """Builder Config for PCBM Metashift"""

    def __init__(
        self,
        metadata_path: str,
        selected_classes: list[str],
        spurious_class: str,
        source_context: str,
        target_context: str,
        **kwargs,
    ):
        super(PCBMMetashiftConfig, self).__init__(
            version=datasets.Version("2.0.3"), **kwargs
        )
        self.metadata_path = metadata_path
        self.selected_classes = selected_classes
        self.spurious_class = spurious_class
        self.source_context = source_context
        self.target_context = target_context


class PCBMMetashift(datasets.GeneratorBasedBuilder):
    """PCBM Metashift dataset"""

    setups = ["cherrypicked", "seed42"]
    BUILDER_CONFIGS = list(
        itertools.chain(
            *[
                [
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_1_bed_cat_dog",
                        description=f"[{setup}] Task 1: bed(cat) -> bed(dog)",
                        metadata_path=f"scenarios/{setup}/task_1_bed_cat_dog.json",
                        selected_classes=["airplane", "bed", "car", "cow", "keyboard"],
                        spurious_class="bed",
                        source_context="cat",
                        target_context="dog",
                    ),
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_1_bed_dog_cat",
                        description=f"[{setup}] Task 1: bed(dog) -> bed(cat)",
                        metadata_path=f"scenarios/{setup}/task_1_bed_dog_cat.json",
                        selected_classes=["airplane", "bed", "car", "cow", "keyboard"],
                        spurious_class="bed",
                        source_context="dog",
                        target_context="cat",
                    ),
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_2_table_books_cat",
                        description=f"[{setup}] Task 2: table(books) -> table(cat)",
                        metadata_path=f"scenarios/{setup}/task_2_table_books_cat.json",
                        selected_classes=[
                            "beach",
                            "computer",
                            "motorcycle",
                            "stove",
                            "table",
                        ],
                        spurious_class="table",
                        source_context="books",
                        target_context="cat",
                    ),
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_2_table_books_dog",
                        description=f"[{setup}] Task 2: table(books) -> table(dog)",
                        metadata_path=f"scenarios/{setup}/task_2_table_books_dog.json",
                        selected_classes=[
                            "beach",
                            "computer",
                            "motorcycle",
                            "stove",
                            "table",
                        ],
                        spurious_class="table",
                        source_context="books",
                        target_context="dog",
                    ),
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_2_table_cat_dog",
                        description=f"[{setup}] Task 2: table(cat) -> table(dog)",
                        metadata_path=f"scenarios/{setup}/task_2_table_cat_dog.json",
                        selected_classes=[
                            "beach",
                            "computer",
                            "motorcycle",
                            "stove",
                            "table",
                        ],
                        spurious_class="table",
                        source_context="cat",
                        target_context="dog",
                    ),
                    PCBMMetashiftConfig(
                        name=f"{setup}_task_2_table_dog_cat",
                        description=f"[{setup}] Task 2: table(dog) -> table(cat)",
                        metadata_path=f"scenarios/{setup}/task_2_table_dog_cat.json",
                        selected_classes=[
                            "beach",
                            "computer",
                            "motorcycle",
                            "stove",
                            "table",
                        ],
                        spurious_class="table",
                        source_context="dog",
                        target_context="cat",
                    ),
                ]
                for setup in setups
            ]
        )
    )

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION_TEMPLATE.substitute(
                num_classes=len(self.config.selected_classes),
                spurious_class=self.config.spurious_class,
                source_context=self.config.source_context,
                target_context=self.config.target_context,
                selected_classes=", ".join(self.config.selected_classes),
            ),
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.ClassLabel(names=self.config.selected_classes),
                }
            ),
            supervised_keys=("image", "label"),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            license=_LICENSE,
            task_templates=[
                datasets.ImageClassification(image_column="image", label_column="label")
            ],
        )

    def _split_generators(self, dl_manager):
        archive_path = dl_manager.download(f"{_REPO}/data/images.tar.gz")
        metadata_path = dl_manager.download(f"{_REPO}/{self.config.metadata_path}")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "images": dl_manager.iter_archive(archive_path),
                    "metadata_path": metadata_path,
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "images": dl_manager.iter_archive(archive_path),
                    "metadata_path": metadata_path,
                    "split": "test",
                },
            ),
        ]

    def _generate_examples(self, images, metadata_path: str, split: str):
        """Generate images and labels for splits."""
        with open(metadata_path, encoding="utf-8") as f:
            metadata = json.load(f)
            split_data = metadata["data_splits"][split]
            ids_to_keep = set()
            for _, ids in split_data.items():
                ids_to_keep.update([Path(id).stem for id in ids])
        files = dict()
        for file_path, file_obj in images:
            image_id = Path(file_path).stem
            if image_id in ids_to_keep:
                files[image_id] = (file_obj.read(), file_path)

        for cls, ids in split_data.items():
            for image_id in ids:
                image_id = Path(image_id).stem
                file_obj, file_path = files[image_id]
                yield f"{cls}_{image_id}", {
                    "image": {"path": file_path, "bytes": file_obj},
                    "label": cls,
                }