renovation.py

import os
import glob
import random

import datasets
from datasets.tasks import ImageClassification

_HOMEPAGE = "https://github.com/your-github/renovation"

_CITATION = """\
@ONLINE {renovationdata,
    author="Your Name",
    title="Renovation dataset",
    month="January",
    year="2023",
    url="https://github.com/your-github/renovation"
}
"""

_DESCRIPTION = """\
Renovations is a dataset of images of houses taken in the field using smartphone
cameras. It consists of 3 classes: cheap, average, and expensive renovations.
Data was collected by the your research lab.
"""

_URLS = {
    'Not Applicable': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Not Applicable.zip",
    'Very Poor': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Very Poor.zip",
    'Poor': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Poor.zip",
    'Fair': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Fair.zip",
    'Good': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Good.zip",
    'Excellent': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Excellent.zip",
    'Exceptional': "https://huggingface.co/datasets/rshrott/photos/resolve/main/Exceptional.zip"
}


_NAMES = ['Not Applicable', 'Very Poor', 'Poor', 'Fair', 'Good', 'Excellent', 'Exceptional']

class Renovations(datasets.GeneratorBasedBuilder):
    """Renovations house images dataset."""

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image_file_path": datasets.Value("string"),
                    "image": datasets.Image(),
                    "labels": datasets.features.ClassLabel(names=_NAMES),
                }
            ),
            supervised_keys=("image", "labels"),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            task_templates=[ImageClassification(image_column="image", label_column="labels")],
        )

    def _split_generators(self, dl_manager):
        data_files = dl_manager.download_and_extract(_URLS)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "data_files": data_files,
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "data_files": data_files,
                    "split": "val",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "data_files": data_files,
                    "split": "test",
                },
            ),
        ]
    def _generate_examples(self, data_files, split):
        all_files_and_labels = []
        for label, path in data_files.items():
            files = glob.glob(path + '/*.jpeg', recursive=True)
            all_files_and_labels.extend((file, label) for file in files)
    
        random.seed(43)  # ensure reproducibility
        random.shuffle(all_files_and_labels)
    
        num_files = len(all_files_and_labels)
        train_data = all_files_and_labels[:int(num_files*0.9)]
        val_test_data = all_files_and_labels[int(num_files*0.9):] # This will be used for both val and test
    
        if split == "train":
            data_to_use = train_data
        else: # "val" or "test" split
            data_to_use = val_test_data
    
        for idx, (file, label) in enumerate(data_to_use):
            yield idx, {
                "image_file_path": file,
                "image": file,
                "labels": label,
        }