SMVB.py

 # 
 # This file is part of the SMVB distribution (https://huggingface.co/datasets/ABC-iRobotics/SMVB).
 # Copyright (c) 2023 ABC-iRobotics.
 # 
 # This program is free software: you can redistribute it and/or modify  
 # it under the terms of the GNU General Public License as published by  
 # the Free Software Foundation, version 3.
 #
 # This program is distributed in the hope that it will be useful, but 
 # WITHOUT ANY WARRANTY; without even the implied warranty of 
 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 
 # General Public License for more details.
 #
 # You should have received a copy of the GNU General Public License 
 # along with this program. If not, see <http://www.gnu.org/licenses/>.
 #
"""SMVB dataset"""

import sys
import io
import numpy as np
if sys.version_info < (3, 9):
    from typing import Sequence, Generator, Tuple
else:
    from collections.abc import Sequence, Generator
    Tuple = tuple

from typing import Optional, IO

import datasets
import itertools
from huggingface_hub import HfFileSystem


# ---- Constants ----

_CITATION = """\
@INPROCEEDINGS{karoly2024synthetic,
  author={Károly, Artúr I. and Nádas, Imre and Galambos, Péter},
  booktitle={2024 IEEE 22nd World Symposium on Applied Machine Intelligence and Informatics (SAMI)}, 
  title={Synthetic Multimodal Video Benchmark (SMVB): Utilizing Blender for rich dataset generation}, 
  year={2024},
  volume={},
  number={},
  pages={},
  doi={}}

"""

_DESCRIPTION = """\
Amultimodal video benchmark for evaluating models in multi-task learning scenarios.
"""

_HOMEPAGE = "https://huggingface.co/ABC-iRobotics/SMVB"

_LICENSE = "GNU General Public License v3.0"

_BASE_URL = "https://huggingface.co/"
_REPO = "datasets/ABC-iRobotics/SMVB"
_RESOURCE = "/resolve/main"

_VERSION = "1.0.0"


# ---- SMVB dataset Configs ----

class SMVBDatasetConfig(datasets.BuilderConfig):
    """BuilderConfig for SMVB dataset."""

    def __init__(self, name: str,  version: Optional[str] = None, **kwargs):
        super(SMVBDatasetConfig, self).__init__(version=datasets.Version(version), name=name, **kwargs)
        fs = HfFileSystem()
        tarfiles = sorted(fs.glob(_REPO + "/**.tar.gz"))
        self._data_urls = [p.replace(_REPO,_BASE_URL+_REPO+_RESOURCE) for p in tarfiles]

    @property
    def features(self):
        return datasets.Features(
            {
                "image": datasets.Image(),
                "mask": datasets.Image(),
                "depth": datasets.Sequence(datasets.Value("float32")),
                "flow": datasets.Sequence(datasets.Value("float32")),
                "normal": datasets.Sequence(datasets.Value("float32"))
            }
        )
    
    @property
    def keys(self):
        return ("image", "mask", "depth", "flow", "normal")



# ---- SMVB dataset Loader ----

class SMVBDataset(datasets.GeneratorBasedBuilder):
    """SMVB dataset."""

    BUILDER_CONFIG_CLASS = SMVBDatasetConfig
    BUILDER_CONFIGS = [
        SMVBDatasetConfig(
            name = "all",
            description = "Synthetic data with rich annotations",
            version = _VERSION
            ),
    ]
    DEFAULT_WRITER_BATCH_SIZE = 10

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=self.config.features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
            version=self.config.version,
        )

    def _split_generators(self, dl_manager):
        local_data_paths = dl_manager.download(self.config._data_urls)
        local_data_gen = itertools.chain.from_iterable([dl_manager.iter_archive(path) for path in local_data_paths])
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "data": local_data_gen
                }
            )
        ]

    def _generate_examples(
        self,
        data: Generator[Tuple[str,IO], None, None]
    ):
        file_infos = []
        keys = self.config.keys

        for i, info in enumerate(data):
            file_path, file_object = info
            if i%len(keys) < 2:
                file_infos.append((file_path, file_object.read()))
            else:
                file_infos.append((file_path, np.load(io.BytesIO(file_object.read())).flatten()))
            if (i+1)%len(keys) == 0:
                img_features_dict = {k:{'path':d[0],'bytes':d[1]} for k,d in zip(keys,file_infos) if k in ['image','mask']}
                array_features_dict = {k:d[1] for k,d in zip(keys,file_infos) if not k in ['image','mask']}
                data_dict = {**img_features_dict, **array_features_dict}
                yield (i//len(keys))-1, data_dict
                file_infos = []