RDD_2020.py

# TODO: Address all TODOs and remove all explanatory comments
"""TODO: Add a description here."""

import csv
import json
import os
from typing import List
import datasets
import logging
import xml.etree.ElementTree as ET
import os

# TODO: Add BibTeX citation
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@InProceedings{huggingface:dataset,
title = {A great new dataset},
author={Shixuan An
},
year={2024}
}
"""

# TODO: Add description of the dataset here
# You can copy an official description
_DESCRIPTION = """\
This new dataset is designed to solve this great NLP task and is crafted with a lot of care.
"""

# TODO: Add a link to an official homepage for the dataset here
_HOMEPAGE = ""

# TODO: Add the licence for the dataset here if you can find it
_LICENSE = ""

# TODO: Add link to the official dataset URLs here
# The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
_URLS = {
    "dataset": "https://prod-dcd-datasets-cache-zipfiles.s3.eu-west-1.amazonaws.com/5ty2wb6gvg-1.zip"
}


# TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
class RDD2020_Dataset(datasets.GeneratorBasedBuilder):
    """TODO: Short description of my dataset."""

    _URLS = _URLS
    VERSION = datasets.Version("1.1.0")

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features({
                "image_id": datasets.Value("string"),
                "country": datasets.Value("string"),
                "type": datasets.Value("string"),
                "image_resolution": datasets.Features({
                    "width": datasets.Value("int32"),
                    "height": datasets.Value("int32"),
                    "depth": datasets.Value("int32"),
                }),
                "image_path": datasets.Value("string"),
                #"pics_array": datasets.Array3D(shape=(None, None, 3), dtype="uint8"),
                "crack_type": datasets.Sequence(datasets.Value("string")),
                "crack_coordinates": datasets.Sequence(datasets.Features({
                    "x_min": datasets.Value("int32"),
                    "x_max": datasets.Value("int32"),
                    "y_min": datasets.Value("int32"),
                    "y_max": datasets.Value("int32"),
                })),
            }),
            homepage='https://data.mendeley.com/datasets/5ty2wb6gvg/1',
            citation=_CITATION,
        )
        
    def _split_generators(self, dl_manager):
        # The direct links to the zipped files on Hugging Face
        urls_to_download = {
            "test2": "https://huggingface.co/datasets/ShixuanAn/RDD_2020/raw/main/test2.zip",
            "test1": "https://huggingface.co/datasets/ShixuanAn/RDD_2020/raw/main/test1.zip",
            "train": "https://huggingface.co/datasets/ShixuanAn/RDD_2020/raw/main/train.zip"
        }

        # Download and extract the dataset using the dl_manager
        downloaded_files = {
            key: dl_manager.download_and_extract(url) for key, url in urls_to_download.items()
        }

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "images_dir": os.path.join(downloaded_files["train"], "images"),
                    "annotations_dir": os.path.join(downloaded_files["train"], "annotations", "xmls"),
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "images_dir": os.path.join(downloaded_files["test1"], "images"),
                    "annotations_dir": None,  # No annotations for test1
                    "split": "test1",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "images_dir": os.path.join(downloaded_files["test2"], "images"),
                    "annotations_dir": None,  # No annotations for test2
                    "split": "test2",
                },
            ),
        ]

    def _generate_examples(self, images_dir, annotations_dir, split):
        # Loop over each country directory in the images_dir
        for country_dir in os.listdir(images_dir):
            country_images_dir = os.path.join(images_dir, country_dir)
            country_annotations_dir = os.path.join(annotations_dir, country_dir, "xmls") if annotations_dir else None

            # Now loop over each image in the country's image directory
            for image_file in os.listdir(country_images_dir):
                if not image_file.endswith('.jpg'):
                    continue
                image_id = image_file.split('.')[0]
                annotation_file = image_id + '.xml'
                annotation_path = os.path.join(country_annotations_dir, annotation_file) if country_annotations_dir else None

                if annotation_path and not os.path.exists(annotation_path):
                    continue

                # Parse the XML file for annotations if it exists
                crack_type = []
                crack_coordinates = []
                if annotation_path:
                    tree = ET.parse(annotation_path)
                    root = tree.getroot()
                    for obj in root.findall('object'):
                        crack_type.append(obj.find('name').text)
                        bndbox = obj.find('bndbox')
                        coordinates = {
                            "x_min": int(bndbox.find('xmin').text),
                            "x_max": int(bndbox.find('xmax').text),
                            "y_min": int(bndbox.find('ymin').text),
                            "y_max": int(bndbox.find('ymax').text),
                        }
                        crack_coordinates.append(coordinates)

                # Assuming images are of uniform size, you might want to adjust this or extract from image directly
                image_resolution = {"width": 600, "height": 600, "depth": 3} if country_dir != "India" else {"width": 720, "height": 720, "depth": 3}

                # Yield the example as a key, value pair
                yield image_id, {
                    "image_id": image_id,
                    "country": country_dir,
                    "type": split,
                    "image_resolution": image_resolution,
                    "image_path": os.path.join(country_images_dir, image_file),
                    "crack_type": crack_type,
                    "crack_coordinates": crack_coordinates,
                }