odor.py

# Based on: https://huggingface.co/datasets/cppe-5/blob/main/cppe-5.py.
"""ODOR dataset."""


import collections
import json
import os
import pandas as pd

import datasets
import time

import requests
import pandas as pd
from tqdm import tqdm
from multiprocessing.pool import ThreadPool
from multiprocessing import cpu_count
from requests.exceptions import MissingSchema, Timeout, ConnectionError


_CITATION = """\
TBD
"""

_DESCRIPTION = """\
Real-world applications of computer vision in the humanities require algorithms to be robust against artistic abstraction, peripheral objects, and subtle differences between fine-grained target classes. Existing datasets provide instance-level
annotations on artworks but are generally biased towards the image centre and limited with regard to detailed object classes. The proposed ODOR dataset fills this gap, offering 38,116 object-level annotations across 4,712 images, spanning an extensive set of 139 fine-grained categories. Conducting a statistical analysis, we showcase challenging dataset properties, such as a detailed set of categories, dense and overlapping objects, and spatial distribution over the whole image canvas. Furthermore, we provide an extensive baseline analysis for object detection models and highlight the challenging properties of the dataset through a set of secondary studies. Inspiring further research on artwork object detection and broader visual cultural heritage studies, the dataset challenges researchers to explore the intersection of object recognition and smell perception.
"""

_HOMEPAGE = "https://zenodo.org/record/10027116"

_LICENSE = "CC BY 4.0"

_META_URL = 'https://zenodo.org/records/10027116/files/meta.csv?download=1'
_TRAIN_JSON_URL = 'https://zenodo.org/records/10027116/files/instances_train.json?download=1'
_TEST_JSON_URL = 'https://zenodo.org/records/10027116/files/instances_test.json?download=1'

_CATEGORIES = ['ant', 'camel', 'jewellery', 'frog', 'physalis', 'celery', 'cauliflower', 'pepper', 'ranunculus', 'chess flower', 'cigarette', 'matthiola', 'cabbage', 'earring', 'dandelion', 'neroli', 'dragonfly', 'hyacinth', 'reptile/amphibia', 'apricot', 'snake', 'lizard', 'asparagus', 'spring onion', 'snowflake', 'moth', 'poppy', 'columbine', 'rabbit', 'geranium', 'crab', 'radish', 'big cat', 'jan steen jug', 'monkey', 'snail', 'bellflower', 'lilac', 'pot', 'peony', 'coffeepot', 'hazelnut', 'censer', 'artichoke', 'dahlia', 'sniffing', 'fly', 'deer', 'caterpillar', 'garlic', 'blackberry', 'chalice', 'lobster', 'necklace', 'bug', 'insect', 'prawn', 'bracelet', 'carrot', 'cornflower', 'pumpkin', 'orange', 'walnut', 'cat', 'daisy', 'forget-me-not', 'carafe', 'match', 'beer stein', 'tobacco-box', 'violet', 'pomander', 'bottle', 'candle', 'heliotrope', 'wine bottle', 'strawberry', 'pomegranate', 'whale', 'lily of the valley', 'iris', 'tobacco', 'olive', 'tobacco-packaging', 'meat', 'daffodil', 'melon', 'fire', 'petunia', 'mushroom', 'teapot', 'ring', 'pig', 'ashtray', 'cheese', 'onion', 'cup', 'nut', 'fig', 'drinking vessel', 'donkey', 'holding the nose', 'lily', 'smoke', 'bread', 'currant', 'glass without stem', 'anemone', 'mammal', 'chimney', 'smoking equipment', 'bivalve', 'butterfly', 'gloves', 'lemon', 'horse', 'plum', 'jasmine', 'pear', 'glass with stem', 'vegetable', 'carnation', 'jug', 'goat', 'fish', 'apple', 'tulip', 'cherry', 'cow', 'animal corpse', 'dog', 'fruit', 'bird', 'rose', 'peach', 'sheep', 'pipe', 'grapes', 'flower']


class ODOR(datasets.GeneratorBasedBuilder):
    """ODOR dataset."""

    VERSION = datasets.Version("0.0.1")

    def _info(self):
        features = datasets.Features(
            {
                "image_id": datasets.Value("int64"),
                "image": datasets.Image(),
                "width": datasets.Value("int32"),
                "height": datasets.Value("int32"),
                "objects": datasets.Sequence(
                    {
                        "id": datasets.Value("int64"),
                        "area": datasets.Value("int64"),
                        "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
                        "category": datasets.ClassLabel(names=_CATEGORIES),
                    }
                ),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):

        def _download_file(uri, target_dir, fn, retries, overwrite=False):
            path = f'{target_dir}/{fn}'

            if os.path.exists(path) and not overwrite:
                return fn

            for i in range(retries):
                try:
                    r = requests.get(uri, stream=True, timeout=50)
                except (MissingSchema, Timeout, ConnectionError):
                    time.sleep(i)
                    continue
                if r.status_code == 200:
                    with open(path, 'wb') as f:
                        for chunk in r:
                            f.write(chunk)
                    return fn
                else:
                    time.sleep(i)
                    continue
            return fn

        def _download_one(entry, overwrite=False):
            fn, uri, target_pth, retries = entry
            fn = fn.replace("/", "_")

            return _download_file(uri, target_pth, fn, retries, overwrite)

        def _download_all(metadata_pth, target_pth, retries=3):
            df = pd.read_csv(metadata_pth)
            entries = [[*x, target_pth, retries] for x in df[['File Name', 'Image Credits']].values]
            n_processes = max(1, cpu_count() - 1)
            with ThreadPool(n_processes) as p:
                results = list(tqdm(p.imap(_download_one, entries), total=len(entries)))
            return results

        imgs_dir = f'{self.cache_dir}/images' 
        _download_file(_META_URL, '.', 'meta.csv', 3)
        if not os.path.isdir('./annotations'):
            os.makedirs('./annotations')
        _download_file(_TRAIN_JSON_URL, './annotations', 'train.json',3)
        _download_file(_TEST_JSON_URL, './annotations', 'test.json',3)
        csv_pth = f'./meta.csv'
        if not os.path.isdir(imgs_dir):
            os.makedirs(imgs_dir)
        
        img_pths = _download_all(csv_pth, imgs_dir)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "annotation_file_path": "annotations/train.json",
                    "metadata_file_path": csv_pth,
                    "img_dir": imgs_dir
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "annotation_file_path": "annotations/test.json",
                    "metadata_file_path": csv_pth,
                    "img_dir": imgs_dir
                },
            ),
        ]

    def _generate_examples(self, annotation_file_path, metadata_file_path, img_dir):

        def process_annot(annot, category_id_to_category):
            return {
                "id": annot["id"],
                "area": annot["area"],
                "bbox": annot["bbox"],
                "category": category_id_to_category[annot["category_id"]],
            }

        image_id_to_image = {}
        idx = 0

        with open(annotation_file_path) as f:
            annotations = json.load(f)
        category_id_to_category = {category["id"]: category["name"] for category in annotations["categories"]}
        image_id_to_annotations = collections.defaultdict(list)
        for annot in annotations["annotations"]:
            image_id_to_annotations[annot["image_id"]].append(annot)
        image_id_to_image = {annot["file_name"]: annot for annot in annotations["images"]}

        for path in os.listdir(img_dir):
            file_name = os.path.basename(path)
            if file_name in image_id_to_image:
                with open(f'{img_dir}/{path}','rb') as f:
                    image = image_id_to_image[file_name]
                    objects = [
                        process_annot(annot, category_id_to_category) for annot in image_id_to_annotations[image["id"]]
                    ]
                    yield idx, {
                        "image_id": image["id"],
                        "image": {"path": path, "bytes": f.read()},
                        "width": image["width"],
                        "height": image["height"],
                        "objects": objects,
                    }
                    idx += 1