renovation.py

import csv
import random
import datasets
import requests
import os

from PIL import Image
from io import BytesIO
from datasets.tasks import ImageClassification

_HOMEPAGE = "https://huggingface.co/datasets/rshrott/renovation"

_CITATION = """\
@ONLINE {renovationquality,
    author="Your Name",
    title="Renovation Quality Dataset",
    month="Your Month",
    year="Your Year",
    url="https://huggingface.co/datasets/rshrott/renovation"
}
"""

_DESCRIPTION = """\
This dataset contains images of various properties, along with labels indicating the quality of renovation - 'cheap', 'average', 'expensive'.
"""

_URL = "https://huggingface.co/datasets/rshrott/renovation/raw/main/labels.csv"

_NAMES = ["cheap", "average", "expensive"]

class RenovationQualityDataset(datasets.GeneratorBasedBuilder):
    """Renovation Quality Dataset."""

    VERSION = datasets.Version("1.0.0")

    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):
        csv_path = dl_manager.download(_URL)
        with open(csv_path, "r") as f:
            reader = csv.reader(f)
            next(reader)  # skip header
            rows = list(reader)
        
        # Shuffle rows
        random.shuffle(rows)

        # 80% for training, 10% for validation, 10% for testing
        train_end = int(0.8 * len(rows))
        val_end = int(0.9 * len(rows))

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "rows": rows[:train_end],
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "rows": rows[train_end:val_end],
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "rows": rows[val_end:],
                },
            ),
        ]

    def _generate_examples(self, rows):
        def url_to_image(url):
            response = requests.get(url)
            img = Image.open(BytesIO(response.content))
            return img
    
        for id_, row in enumerate(rows):
            if len(row) < 2:
                print(f"Row with id {id_} has less than 2 elements: {row}")
            else:
                image_file_path = str(row[0])
                image = url_to_image(image_file_path)
                yield id_, {
                    'image_file_path': image_file_path,
                    'image': image,
                    'labels': row[1],
                }