ANAKIN.py

import csv
import json
import os
import random

import datasets
import pandas as pd
from PIL import Image
from torchvision.io import read_video

# TODO: Add BibTeX citation
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@misc{black2023vader,
      title={VADER: Video Alignment Differencing and Retrieval}, 
      author={Alexander Black and Simon Jenni and Tu Bui and Md. Mehrab Tanjim and Stefano Petrangeli and Ritwik Sinha and Viswanathan Swaminathan and John Collomosse},
      year={2023},
      eprint={2303.13193},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
"""

# TODO: Add description of the dataset here
# You can copy an official description
_DESCRIPTION = """\
ANAKIN is a dataset of mANipulated videos and mAsK annotatIoNs.
"""

# TODO: Add a link to an official homepage for the dataset here
_HOMEPAGE = "https://github.com/AlexBlck/vader"

# TODO: Add the licence for the dataset here if you can find it
_LICENSE = "cc-by-4.0"

# 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)
_METADATA_URL = "https://huggingface.co/datasets/AlexBlck/ANAKIN/raw/main/metadata.csv"

_FOLDERS = {
    "all": ("full", "trimmed", "edited", "masks"),
    "no-full": ("trimmed", "edited", "masks"),
    "trimmed-edited-masks": ("trimmed", "edited", "masks"),
    "full-trimmed-edited-masks": ("full", "trimmed", "edited", "masks"),
}


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

    VERSION = datasets.Version("1.0.0")

    # 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="all",
            version=VERSION,
            description="Full video, trimmed video, edited video, masks (if exists), and edit description",
        ),
        datasets.BuilderConfig(
            name="no-full",
            version=VERSION,
            description="Trimmed video, edited video, masks (if exists), and edit description",
        ),
        datasets.BuilderConfig(
            name="trimmed-edited-masks",
            version=VERSION,
            description="Only samples that have masks. Without full length video.",
        ),
        datasets.BuilderConfig(
            name="full-trimmed-edited-masks",
            version=VERSION,
            description="Only samples that have masks. With full length video.",
        ),
    ]

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

    def _info(self):
        # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
        if self.config.name == "all":
            features = datasets.Features(
                {
                    "full": datasets.Value("string"),
                    "trimmed": datasets.Value("string"),
                    "edited": datasets.Value("string"),
                    "masks": datasets.Sequence(datasets.Image()),
                    "task": datasets.Value("string"),
                    "start-time": datasets.Value("int32"),
                    "end-time": datasets.Value("int32"),
                    "manipulation-type": datasets.Value("string"),
                    "editor-id": datasets.Value("string"),
                }
            )
        elif self.config.name == "no-full":
            features = datasets.Features(
                {
                    "trimmed": datasets.Value("string"),
                    "edited": datasets.Value("string"),
                    "masks": datasets.Sequence(datasets.Image()),
                    "task": datasets.Value("string"),
                    "manipulation-type": datasets.Value("string"),
                    "editor-id": datasets.Value("string"),
                }
            )
        elif self.config.name == "trimmed-edited-masks":
            features = datasets.Features(
                {
                    "trimmed": datasets.Value("string"),
                    "edited": datasets.Value("string"),
                    "masks": datasets.Sequence(datasets.Image()),
                    "task": datasets.Value("string"),
                    "manipulation-type": datasets.Value("string"),
                    "editor-id": datasets.Value("string"),
                }
            )
        elif self.config.name == "full-trimmed-edited-masks":
            features = datasets.Features(
                {
                    "full": datasets.Value("string"),
                    "trimmed": datasets.Value("string"),
                    "edited": datasets.Value("string"),
                    "masks": datasets.Sequence(datasets.Image()),
                    "task": datasets.Value("string"),
                    "start-time": datasets.Value("int32"),
                    "end-time": datasets.Value("int32"),
                    "manipulation-type": datasets.Value("string"),
                    "editor-id": datasets.Value("string"),
                }
            )
        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
            # If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
            # specify them. They'll be used if as_supervised=True in builder.as_dataset.
            # supervised_keys=("sentence", "label"),
            # Homepage of the dataset for documentation
            homepage=_HOMEPAGE,
            # License for the dataset if available
            license=_LICENSE,
            # Citation for the dataset
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        # TODO: This method is tasked with downloading/extracting the data and defining the splits depending on the configuration
        # If several configurations are possible (listed in BUILDER_CONFIGS), the configuration selected by the user is in self.config.name

        # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
        # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
        # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
        metadata_dir = dl_manager.download(_METADATA_URL)
        folders = _FOLDERS[self.config.name]

        random.seed(47)
        root_url = "https://huggingface.co/datasets/AlexBlck/ANAKIN/resolve/main/"
        df = pd.read_csv(metadata_dir)
        if "full" in folders:
            df = df[df["full-available"] == True]
        if "masks" in folders:
            df = df[df["has-masks"] == True]

        ids = df["video-id"].to_list()
        random.shuffle(ids)

        train_end = int(len(df) * 0.7)
        val_end = int(len(df) * 0.8)
        split_ids = {
            datasets.Split.TRAIN: ids[:train_end],
            datasets.Split.VALIDATION: ids[train_end:val_end],
            datasets.Split.TEST: ids[val_end:],
        }

        data_dir = {}
        mask_dir = {}

        for split in [
            datasets.Split.TRAIN,
            datasets.Split.VALIDATION,
            datasets.Split.TEST,
        ]:
            data_urls = [
                {
                    f"{folder}": root_url + f"{folder}/{idx}.mp4"
                    for folder in folders
                    if folder != "masks"
                }
                for idx in split_ids[split]
            ]
            data_dir[split] = dl_manager.download(data_urls)
            mask_dir[split] = {
                idx: dl_manager.iter_archive(
                    dl_manager.download(root_url + f"masks/{idx}.zip")
                )
                for idx in split_ids[split]
            }

        return [
            datasets.SplitGenerator(
                name=split,
                gen_kwargs={
                    "files": data_dir[split],
                    "masks": mask_dir[split],
                    "df": df,
                    "ids": split_ids[split],
                    "return_time": "full" in folders,
                },
            )
            for split in [
                datasets.Split.TRAIN,
                datasets.Split.VALIDATION,
                datasets.Split.TEST,
            ]
        ]

    def _generate_examples(self, files, masks, df, ids, return_time):
        for key, (idx, sample) in enumerate(zip(ids, files)):
            print(idx)
            entry = df[df["video-id"] == idx]
            print(entry)
            if entry["has-masks"].values[0]:
                sample["masks"] = [
                    {"path": p, "bytes": im.read()} for p, im in masks[idx]
                ]
            else:
                sample["masks"] = None
            sample["task"] = entry["task"].values[0]
            sample["manipulation-type"] = entry["manipulation-type"].values[0]
            sample["editor-id"] = entry["editor-id"].values[0]
            if return_time:
                sample["start-time"] = entry["start-time"].values[0]
                sample["end-time"] = entry["end-time"].values[0]
            yield key, sample