ecoset.py

import sys
import csv
import json
import os
import os.path as op
import base64
import zipfile
from getpass import getpass
from tqdm import tqdm
import platform
import subprocess

from urllib.parse import urlparse
import datasets
from datasets.filesystems import S3FileSystem
import boto3
from botocore import UNSIGNED
from botocore.client import Config


# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@article{mehrer2021ecologically,
  title={An ecologically motivated image dataset for deep learning yields better models of human vision},
  author={Mehrer, Johannes and Spoerer, Courtney J and Jones, Emer C and Kriegeskorte, Nikolaus and Kietzmann, Tim C},
  journal={Proceedings of the National Academy of Sciences},
  volume={118},
  number={8},
  year={2021},
  publisher={National Acad Sciences}
}
"""

# You can copy an official description
_DESCRIPTION = """\
Tired of all the dogs in ImageNet (ILSVRC)? Then ecoset is here for you. 1.5m images 
from 565 basic level categories, chosen to be both (i) frequent in linguistic usage, 
and (ii) rated by human observers as concrete (e.g. ‘table’ is concrete, ‘romance’ 
is not). Here we collect resources associated with ecoset. This includes the dataset,
trained deep neural network models, code to interact with them, and published papers 
using it.
"""

# official homepage for the dataset here
_HOMEPAGE = "https://www.kietzmannlab.org/ecoset/"

# licence for the dataset here
_LICENSE = "CC BY NC SA 2.0"

# 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 = {
    #"codeocean": "https://files.codeocean.com/datasets/verified/0ab003f4-ff2d-4de3-b4f8-b6e349c0e5e5/ecoset.zip?download",  # codeocean cancels after 50GB
    "codeocean": "s3://codeocean-datasets/0ab003f4-ff2d-4de3-b4f8-b6e349c0e5e5/ecoset.zip",
}

# Define the labels available for ecoset
labels = ['cymbals', 'bison', 'lemonade', 'crib', 'chestnut', 'mosquito', 'aloe', 'extinguisher', 'onion', 'starfish', 'basket', 'jar', 'snail', 'mushroom', 'coffin', 'joystick', 'raspberry', 'gearshift', 'tyrannosaurus', 'stadium', 'telescope', 'blueberry', 'hippo', 'cannabis', 'hairbrush', 'river', 'artichoke', 'wallet', 'city', 'bee', 'rifle', 'boar', 'bib', 'envelope', 'silverfish', 'shower', 'curtain', 'pinwheel', 'guillotine', 'snowplow', 'hut', 'jukebox', 'gecko', 'marshmallow', 'lobster', 'flashlight', 'breadfruit', 'cow', 'spoon', 'blender', 'croissant', 'greenhouse', 'church', 'antenna', 'monkey', 'zucchini', 'snake', 'manatee', 'child', 'table', 'winterberry', 'sloth', 'cannon', 'baguette', 'persimmon', 'candelabra', 'necklace', 'flag', 'geyser', 'thermos', 'tweezers', 'chandelier', 'kebab', 'mailbox', 'steamroller', 'crayon', 'lawnmower', 'pomegranate', 'fire', 'violin', 'matchstick', 'train', 'hamster', 'bobsleigh', 'boat', 'bullet', 'forklift', 'clock', 'saltshaker', 'anteater', 'crowbar', 'lightbulb', 'pier', 'muffin', 'paintbrush', 'crawfish', 'bench', 'nectarine', 'eyedropper', 'backpack', 'goat', 'hotplate', 'fishnet', 'robot', 'rice', 'shovel', 'candle', 'blimp', 'bridge', 'mountain', 'coleslaw', 'stagecoach', 'waterfall', 'ladle', 'radiator', 'drain', 'tray', 'house', 'key', 'skunk', 'lake', 'earpiece', 'gazebo', 'blackberry', 'groundhog', 'paperclip', 'cookie', 'milk', 'rug', 'thermostat', 'milkshake', 'scoreboard', 'bean', 'giraffe', 'antelope', 'newsstand', 'camcorder', 'sawmill', 'balloon', 'ladder', 'videotape', 'microphone', 'coin', 'hay', 'moth', 'octopus', 'honeycomb', 'wrench', 'cane', 'bobcat', 'banner', 'newspaper', 'reef', 'worm', 'cucumber', 'beach', 'couch', 'streetlamp', 'rhino', 'ceiling', 'cupcake', 'hourglass', 'caterpillar', 'tamale', 'asparagus', 'flower', 'frog', 'dog', 'knife', 'lamp', 'walnut', 'grape', 'scone', 'peanut', 'ferret', 'kettle', 'elephant', 'oscilloscope', 'weasel', 'guava', 'gramophone', 'stove', 'bamboo', 'chicken', 'guacamole', 'toolbox', 'tractor', 'tiger', 'butterfly', 'coffeepot', 'bus', 'meteorite', 'fish', 'graveyard', 'blowtorch', 'grapefruit', 'cat', 'jellyfish', 'carousel', 'wheat', 'tadpole', 'kazoo', 'raccoon', 'typewriter', 'scissors', 'pothole', 'earring', 'drawers', 'cup', 'warthog', 'wall', 'lighthouse', 'burrito', 'cassette', 'nacho', 'sink', 'seashell', 'bed', 'noodles', 'woman', 'rabbit', 'fence', 'pistachio', 'pencil', 'hotdog', 'ball', 'ship', 'strawberry', 'pan', 'custard', 'dolphin', 'tent', 'bun', 'tortilla', 'tumbleweed', 'playground', 'scallion', 'anchor', 'hare', 'waterspout', 'dough', 'burner', 'kale', 'razor', 'chocolate', 'doughnut', 'squeegee', 'bandage', 'beaver', 'refrigerator', 'cork', 'anvil', 'microchip', 'banana', 'thumbtack', 'chair', 'sharpener', 'bird', 'castle', 'wand', 'doormat', 'celery', 'steak', 'ant', 'apple', 'cave', 'scaffolding', 'bell', 'towel', 'mantis', 'thimble', 'bowl', 'chess', 'pickle', 'lollypop', 'leek', 'barrel', 'dollhouse', 'tapioca', 'spareribs', 'fig', 'apricot', 'strongbox', 'brownie', 'beaker', 'manhole', 'piano', 'whale', 'hammer', 'dishrag', 'pecan', 'highlighter', 'pretzel', 'earwig', 'cogwheel', 'trashcan', 'syringe', 'turnip', 'pear', 'lettuce', 'hedgehog', 'guardrail', 'bubble', 'pineapple', 'burlap', 'moon', 'spider', 'fern', 'binoculars', 'gravel', 'plum', 'scorpion', 'cube', 'squirrel', 'book', 'crouton', 'bag', 'lantern', 'parsley', 'jaguar', 'thyme', 'oyster', 'kumquat', 'chinchilla', 'cherry', 'umbrella', 'bicycle', 'eggbeater', 'pig', 'kitchen', 'fondue', 'treadmill', 'casket', 'papaya', 'beetle', 'shredder', 'grasshopper', 'anthill', 'chili', 'bottle', 'calculator', 'gondola', 'pizza', 'compass', 'mop', 'hamburger', 'chipmunk', 'bagel', 'outhouse', 'pliers', 'wolf', 'matchbook', 'corn', 'salamander', 'lasagna', 'stethoscope', 'eggroll', 'avocado', 'eggplant', 'mouse', 'walrus', 'sprinkler', 'glass', 'cauldron', 'parsnip', 'canoe', 'pancake', 'koala', 'deer', 'chalk', 'urinal', 'toilet', 'cabbage', 'platypus', 'lizard', 'leopard', 'cake', 'hammock', 'defibrillator', 'sundial', 'beet', 'popcorn', 'spinach', 'cauliflower', 'canyon', 'spacecraft', 'teapot', 'tunnel', 'porcupine', 'jail', 'spearmint', 'dustpan', 'calipers', 'toast', 'drum', 'phone', 'wire', 'alligator', 'vase', 'motorcycle', 'toothpick', 'coconut', 'lion', 'turtle', 'cheetah', 'bugle', 'casino', 'fountain', 'pie', 'bread', 'meatball', 'windmill', 'gun', 'projector', 'chameleon', 'tomato', 'nutmeg', 'plate', 'bulldozer', 'camel', 'sphinx', 'mall', 'hanger', 'ukulele', 'wheelbarrow', 'ring', 'dildo', 'loudspeaker', 'odometer', 'ruler', 'mousetrap', 'breadbox', 'parachute', 'bolt', 'bracelet', 'library', 'otter', 'airplane', 'pea', 'tongs', 'cactus', 'knot', 'shrimp', 'computer', 'sheep', 'television', 'melon', 'kangaroo', 'helicopter', 'birdcage', 'pumpkin', 'dishwasher', 'crocodile', 'stairs', 'garlic', 'barnacle', 'crate', 'lime', 'axe', 'hairpin', 'egg', 'emerald', 'candy', 'stegosaurus', 'broom', 'mistletoe', 'submarine', 'fireworks', 'peach', 'ape', 'chalkboard', 'bumblebee', 'potato', 'battery', 'guitar', 'opossum', 'volcano', 'llama', 'ashtray', 'sieve', 'coliseum', 'cinnamon', 'moose', 'tree', 'donkey', 'wasp', 'corkscrew', 'gargoyle', 'taco', 'macadamia', 'camera', 'mandolin', 'kite', 'cranberry', 'thermometer', 'tofu', 'closet', 'hovercraft', 'escalator', 'horseshoe', 'wristwatch', 'lemon', 'sushi', 'rat', 'rainbow', 'pillow', 'radish', 'granola', 'okra', 'pastry', 'mango', 'dragonfly', 'flashbulb', 'chalice', 'acorn', 'birdhouse', 'gooseberry', 'locker', 'padlock', 'missile', 'clarinet', 'panda', 'iceberg', 'road', 'flea', 'hazelnut', 'cockroach', 'needle', 'omelet', 'desert', 'condom', 'graffiti', 'iguana', 'bucket', 'photocopier', 'blanket', 'microscope', 'horse', 'nest', 'screwdriver', 'toaster', 'car', 'doll', 'salsa', 'man', 'zebra', 'stapler', 'grate', 'truck', 'bear', 'carrot', 'auditorium', 'cashew', 'shield', 'crown', 'altar', 'pudding', 'cheese', 'rhubarb', 'broccoli', 'tower', 'cumin', 'elevator', 'wheelchair', 'flyswatter']

# handle password entry
_PWD_MSG = "\nIn order to use ecoset, please read the README and License agreement found under:\nhttps://codeocean.com/capsule/9570390\nand enter the mentioned password.\n\nPlease Enter Password:\n"
def check_pass(pw):
    if base64.b64encode(pw.encode("ascii")) != (b"ZWNvc2V0X21zamtr"):
        raise AttributeError("Wrong password! Please try again.")
    else:
        print("Password correct.\n")

# Name of the dataset usually match the script name with CamelCase instead of snake_case
class Ecoset(datasets.GeneratorBasedBuilder):
    """Ecoset is a large clean and ecologically valid image dataset."""

    VERSION = datasets.Version("1.1.0")

    # This is an example of a dataset with multiple configurations.
    # If you don't want/need to define several sub-sets in your dataset,
    # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.

    # If you need to make complex sub-parts in the datasets with configurable options
    # You can create your own builder configuration class to store attribute, inheriting from datasets.BuilderConfig
    # BUILDER_CONFIG_CLASS = MyBuilderConfig

    # You will be able to load one or the other configurations in the following list with
    # data = datasets.load_dataset('my_dataset', 'first_domain')
    # data = datasets.load_dataset('my_dataset', 'second_domain')
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="Full", version=VERSION, description="We could do different splits of the dataset here. But we don't"),
    ]

    DEFAULT_CONFIG_NAME = "Full"  # It's not mandatory to have a default configuration. Just use one if it make sense.

    def _info(self):
        
        # define dataset features
        features=datasets.Features(
            {
                "image": datasets.Image(),
                "label": datasets.ClassLabel(names=labels),
            }
        )
        return datasets.DatasetInfo(
            # This is the description that will appear on the datasets page.
            description=_DESCRIPTION,
            # This defines the different columns of the dataset and their types
            features=features,  # Here we define them above because they are different between the two configurations
            # Homepage of the dataset for documentation
            homepage=_HOMEPAGE,
            # License for the dataset if available
            license=_LICENSE,
            # Citation for the dataset
            citation=_CITATION,
            # creating a template
            task_templates=[datasets.tasks.ImageClassification(image_column="image", label_column="label")],
        )
    

    def _split_generators(self, dl_manager):
        # Ask password of user. This could be also handled through dataset config
        
        def abslist(path):
            """Helper function to give abspaths of os.listdir"""
            return [op.join(path, p) for p in os.listdir(path)]
            
        def subprocess_call_print(command_list):
            """Execute a subprocess while printing the command line output"""
            p = subprocess.Popen(command_list, stdout=subprocess.PIPE)
            while True:
                line = p.stdout.readline()
                if not line:
                    break
                print(line.strip())
                sys.stdout.flush()
        
        def s3_zipfile_download(source_url, target_dir):
            """Extremely slow download"""
            # ask password
            password = getpass(_PWD_MSG)
            check_pass(password)
            # download and unzip
            print('Using slow Python-based download and unzipping. This can take up to 70h on a typical computer. Sorry.')
            s3 = S3FileSystem(anon=True, use_ssl=False, default_block_size=int(15 * 2**20)) 
            with s3.open(source_url, "rb") as raw_filw:
                with ZipFile(raw_filw, compression=zipfile.ZIP_DEFLATED, allowZip64=True) as zip_file:
                    member_list = zip_file.namelist()
                    for member in tqdm(member_list, total=len(member_list), desc="Extracting ecoset to disc"):
                        zip_file.extract(member, target_dir, pwd=password.encode("ascii"))
                        
                        
        def subprocess_download(source_url, target_dir):
            """Moderately slow download"""
            # ask password
            password = getpass(_PWD_MSG)
            check_pass(password)
            # download
            print('Using native OS unzipping. This will take about 15h on a typical Linux/Mac and 8h on a typical Windows Computer.')
            urlinfo = urlparse(source_url, allow_fragments=False)
            # create destination path if not existing
            if not op.exists(target_dir):
                os.makedirs(target_dir)
            # download zip file if not existing  
            zip_path = op.join(target_dir, "ecoset.zip")
            if not op.exists(zip_path):
                s3 = boto3.client(urlinfo.scheme, config=Config(signature_version=UNSIGNED))
                object_size = s3.head_object()["ContentLength"]
                with tqdm.tqdm(total=object_size, unit="B", unit_scale=True, desc=filename) as pbar:
                    s3.download_file(Bucket=urlinfo.netloc, Key=urlinfo.path[1:], Filename=zip_path,
                    Callback=lambda bytes_transferred: pbar.update(bytes_transferred))
            # unzip using platform-based subprocess
            if platform.system() in ("Linux", "Darwin"):
                #subprocess.call(["unzip", "-n", "-P", password.encode("ascii"), "-o", zip_path, "-d", target_dir], shell=False)
                subprocess_call_print(["unzip", "-n", "-P", password.encode("ascii"), "-o", zip_path, "-d", target_dir])
            else:
                #subprocess.call(["tar.exe", "-xf", zip_path, "-C", target_dir, "--passphrase", password], shell=False)
                subprocess_call_print(["tar.exe", "-xf", zip_path, "-C", target_dir, "--passphrase", password])


        def subprocess_download2(source_url, target_dir):
            """Moderately slow download"""
            # ask password
            password = getpass(_PWD_MSG)
            check_pass(password)
            # download
            print('Using native OS unzipping. This will take about 15h on a typical Linux/Mac and 8h on a typical Windows Computer.')
            urlinfo = urlparse(source_url, allow_fragments=False)
            # create destination path if not existing
            if not op.exists(target_dir):
                os.makedirs(target_dir)
            # download zip file if not existing  
            zip_path = op.join(target_dir, "ecoset.zip")
            if not op.exists(zip_path):
                s3 = boto3.client(urlinfo.scheme, config=Config(signature_version=UNSIGNED))
                s3.download_file(urlinfo.netloc, urlinfo.path[1:], zip_path)
            # unzip using platform-based subprocess
            if platform.system() in ("Linux", "Darwin"):
                subprocess.call(["unzip", "-n", "-P", password.encode("ascii"), "-o", zip_path, "-d", target_dir], shell=False)
            else:
                subprocess.call(["tar.exe", "-xf", zip_path, "-C", target_dir, "--passphrase", password], shell=False)
            
            
        # download and unzip using subprocess. S3 download was discontinued due to being extremely slow
        archives = dl_manager.download_custom(_URLS["codeocean"], subprocess_download)
        print("Ecoset files are stored under: \n", archives)
            
        # create a dict containing all files
        split_dict = {split:[] for split in ("train", "val", "test")}
        for split in split_dict.keys():
            fnames = abslist(op.join(archives, split))
            for f in fnames:
                split_dict[split].extend(abslist(f))
        
        # return data splits
        return [datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "archives": split_dict["train"],
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "archives": split_dict["val"],
                    "split": "validation",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "archives": split_dict["test"],
                    "split": "test",
                },
            ),
        ]

    # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
    def _generate_examples(self, archives, split):
        """Yields examples."""
        idx = 0
        for archive in archives:
            if any(archive.endswith(i) for i in (".JPEG", ".JPG", ".jpeg", ".jpg")):
                
                # extract file, label, etc
                file = open(archive, 'rb')
                synset_id, label = archive.split("/")[-2].split("_")
                ex = {"image": {"path": archive, "bytes": file.read()}, "label": label}
                
                yield idx, ex
                idx += 1