Superimposed-Masked-Dataset.py

# coding=utf-8
# Copyright 2022 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.

# This script was modified from the imagenet-1k HF dataset repo

import os

import datasets
from datasets.tasks import ImageClassification

from .classes import IMAGENET2012_CLASSES


_CITATION = """\
@misc{lee2023hardwiring,
      title={Hardwiring ViT Patch Selectivity into CNNs using Patch Mixing}, 
      author={Ariel N. Lee and Sarah Adel Bargal and Janavi Kasera and Stan Sclaroff and Kate Saenko and Nataniel Ruiz},
      year={2023},
      eprint={2306.17848},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
"""

_HOMEPAGE = "https://arielnlee.github.io/PatchMixing/"

_DESCRIPTION = """\
SMD is an occluded ImageNet-1K validation set, created to be an additional way to evaluate the impact of occlusion on model performance. This experiment used a variety of occluder objects that are not in the ImageNet-1K label space and are unambiguous in relationship to objects that reside in the label space.
"""

_DATA_URL = {
    "smd": [
        f"https://huggingface.co/datasets/ariellee/Superimposed-Masked-Dataset/resolve/main/smd_{i}.tar.gz"
        for i in range(1, 41)
    ]
}


class SMD(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")

    DEFAULT_WRITER_BATCH_SIZE = 1000

    def _info(self):
        assert len(IMAGENET2012_CLASSES) == 1000
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.ClassLabel(names=list(IMAGENET2012_CLASSES.values())),
                }
            ),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            task_templates=[ImageClassification(image_column="image", label_column="label")],
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        archives = dl_manager.download(_DATA_URL)

        return [
            datasets.SplitGenerator(
                name="SMD", # "SMD (occluded IN-1K val set)"
                gen_kwargs={
                    "archives": [dl_manager.iter_archive(archive) for archive in archives["smd"]],
                },
            ),
        ]


    def _generate_examples(self, archives):
        """Yields examples."""
        idx = 0
        for archive in archives:
            for path, file in archive:
                if path.endswith(".png"):
                    synset_id = os.path.basename(os.path.dirname(path))
                    label = IMAGENET2012_CLASSES[synset_id]
                    ex = {"image": {"path": path, "bytes": file.read()}, "label": label}
                    yield idx, ex
                    idx += 1