arch-max.py

# Copyright (C) 2022, François-Guillaume Fernandez.

# This program is licensed under the Apache License 2.0.
# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0> for full license details.

"""Arch-MAX (Architecture: Modular Artificial Explainable) dataset."""

import os
import json
from pathlib import Path

import pandas as pd
import datasets

_HOMEPAGE = "https://huggingface.co/datasets/bruno-cotrim/arch-max"

_LICENSE = "Apache License 2.0"

_CITATION = """ """

_DESCRIPTION = """ """

_REPO = "https://huggingface.co/datasets/bruno-cotrim/arch-max/resolve/main"


class HouseXMAConfig(datasets.BuilderConfig):
    """BuilderConfig for HouseXMA."""

    def __init__(self, data_url, metadata_urls, **kwargs):
        """BuilderConfig for Imagette.
        Args:
          data_url: `string`, url to download the zip file from.
          matadata_urls: dictionary with keys 'train' and 'validation' containing the archive metadata URLs
          **kwargs: keyword arguments forwarded to super.
        """
        super(HouseXMAConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
        self.data_url = data_url
        self.metadata_urls = metadata_urls


_COLLUMNS = [
    'Residential',
    'Commercial',
    'Industrial',
    'Cafe',
    'Hotel',
    'Restaurant',
    'Store',
    'MiscCommercial',
    'Suburban',
    'MiscResidential',
    'CountryHouse',
    'ConstructionSite',
    'MiscIndustrial',
    'PowerPlant',
    'WaterTreatment',
    'Door',
    'Window',
    'Awning',
    'Billboard',
    'Porch',
    'Sign',
    'Table',
    'TiledRoof',
    'TiledRoofTop',
    'TiledRoofBottom',
    'VendingMachine',
    'WallSign',
    'Statue',
    'Chimney',
    'Pipe',
    'Machine',
    'Truck',
    'Car',
]


class HouseXMA(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [
        HouseXMAConfig(
            name="Difficulty lvl. 1",
            description="Simplest version of the dataset. All examples show the same textures and camera position.",
            data_url=f"{_REPO}/images/c1.zip",
            metadata_urls=f"{_REPO}/labels/c1/all.csv",
        ),
        HouseXMAConfig(
            name="Difficulty lvl. 2",
            description="Textures vary, camera is static.",
            data_url=f"{_REPO}/images/c2.zip",
            metadata_urls=f"{_REPO}/labels/c2/all.csv",
        ),
        HouseXMAConfig(
            name="Difficulty lvl. 3",
            description="Textures vary, camera varies lightly.",
            data_url=f"{_REPO}/images/c3.zip",
            metadata_urls=f"{_REPO}/labels/c3/all.csv",
        ),
        HouseXMAConfig(
            name="Difficulty lvl. 4",
            description="Textures vary, camera varies lightly, background objects present.",
            data_url=f"{_REPO}/images/c4.zip",
            metadata_urls=f"{_REPO}/labels/c4/all.csv",
        ),
        HouseXMAConfig(
            name="Difficulty lvl. 5",
            description="Textures vary, camera varies heavily, background objects present.",
            data_url=f"{_REPO}/images/c5.zip",
            metadata_urls=f"{_REPO}/labels/c5/all.csv",
        ),
        HouseXMAConfig(
            name="Difficulty lvl. 6",
            description="Textures vary, camera varies heavily, background objects present, brightness and contrast vary.",
            data_url=f"{_REPO}/images/c6.zip",
            metadata_urls=f"{_REPO}/labels/c6/all.csv",
        ),
    ]

    def _info(self):

        features = {
            "image": datasets.Image()
        }
        for col in _COLLUMNS:
            features[col] = datasets.ClassLabel(num_classes=2)

        return datasets.DatasetInfo(
            description=_DESCRIPTION + self.config.description,
            features=datasets.Features(features),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        images_path = dl_manager.download_and_extract(self.config.data_url)
        labels_path = dl_manager.download(self.config.metadata_urls)
        return [
            datasets.SplitGenerator(
                name='all',
                gen_kwargs={
                    "images": images_path,
                    "labels": labels_path,
                },
            )
        ]

    def _generate_examples(self, images, labels):
        idx = 0
        labels = pd.read_csv(labels)

        for ix in range(len(labels)):
            row = labels.iloc[ix].to_dict()
            im_path = os.path.join(images, str(row['id']) + '.jpg')
       

            example = {
                "image": im_path
            }
            for col in _COLLUMNS:
                example[col] = row[col]

            yield idx, example
            idx += 1