oe_dataset.py

 # 
 # This file is part of the oe_dataset distribution (https://huggingface.co/datasets/ABC-iRobotics/oe_dataset).
 # 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/>.
 #
"""OE dataset"""

import sys
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


# ---- Constants ----

_CITATION = """\
@ARTICLE{10145828,
  author={Károly, Artúr István and Tirczka, Sebestyén and Gao, Huijun and Rudas, Imre J. and Galambos, Péter},
  journal={IEEE Transactions on Cybernetics}, 
  title={Increasing the Robustness of Deep Learning Models for Object Segmentation: A Framework for Blending Automatically Annotated Real and Synthetic Data}, 
  year={2023},
  volume={},
  number={},
  pages={1-14},
  doi={10.1109/TCYB.2023.3276485}}

"""

_DESCRIPTION = """\
An instance segmentation dataset for robotic manipulation in a tabletop environment.
The dataset incorporates real and synthetic images for testing sim-to-real model transfer after fine-tuning.
"""

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

_LICENSE = "GNU General Public License v3.0"

_LATEST_VERSIONS = {
    "all": "1.0.0",
    "real": "1.0.0",
    "synthetic": "1.0.0",
    "photoreal": "1.0.0",
    "random": "1.0.0",
}



# ---- OE dataset Configs ----

class OEDatasetConfig(datasets.BuilderConfig):
    """BuilderConfig for OE dataset."""

    def __init__(self, name: str, imgs_urls: Sequence[str], masks_urls: Sequence[str],  version: Optional[str] = None, **kwargs):
        _version = _LATEST_VERSIONS[name] if version is None else version
        _name = f"{name}_v{_version}"
        super(OEDatasetConfig, self).__init__(version=datasets.Version(_version), name=_name, **kwargs)
        self._imgs_urls = {"train": [url + "/train.tar.gz" for url in imgs_urls], "val": [url + "/val.tar.gz" for url in imgs_urls]}
        self._masks_urls = {"train": [url + "/train.tar.gz" for url in masks_urls], "val": [url + "/val.tar.gz" for url in masks_urls]}

    @property
    def features(self):
        return datasets.Features(
            {
                "image": datasets.Image(),
                "mask": datasets.Image(),
            }
        )
    
    @property
    def supervised_keys(self):
        return ("image", "mask")



# ---- OE dataset Loader ----

class OEDataset(datasets.GeneratorBasedBuilder):
    """OE dataset."""

    BUILDER_CONFIG_CLASS = OEDatasetConfig
    BUILDER_CONFIGS = [
        OEDatasetConfig(
            name = "photoreal",
            description = "Photorealistic synthetic images",
            imgs_urls = ["https://huggingface.co/datasets/ABC-iRobotics/oe_dataset/resolve/main/synthetic/photoreal/imgs"],
            masks_urls = ["https://huggingface.co/datasets/ABC-iRobotics/oe_dataset/resolve/main/synthetic/photoreal/masks"]
            ),
            OEDatasetConfig(
            name = "random",
            description = "Domain randomized synthetic images",
            imgs_urls = ["https://huggingface.co/datasets/ABC-iRobotics/oe_dataset/resolve/main/synthetic/random/imgs"],
            masks_urls = ["https://huggingface.co/datasets/ABC-iRobotics/oe_dataset/resolve/main/synthetic/random/masks"]
            ),
    ]

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

    def _split_generators(self, dl_manager):
        train_imgs_paths = dl_manager.download(self.config._imgs_urls["train"])
        val_imgs_paths = dl_manager.download(self.config._imgs_urls["val"])

        train_masks_paths = dl_manager.download(self.config._masks_urls["train"])
        val_masks_paths = dl_manager.download(self.config._masks_urls["val"])

        train_imgs_gen = itertools.chain.from_iterable([dl_manager.iter_archive(path) for path in train_imgs_paths])
        val_imgs_gen = itertools.chain.from_iterable([dl_manager.iter_archive(path) for path in val_imgs_paths])

        train_masks_gen = itertools.chain.from_iterable([dl_manager.iter_archive(path) for path in train_masks_paths])
        val_masks_gen = itertools.chain.from_iterable([dl_manager.iter_archive(path) for path in val_masks_paths])
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "images": train_imgs_gen,
                    "masks": train_masks_gen,
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "images": val_imgs_gen,
                    "masks": val_masks_gen,
                },
            ),
        ]

    def _generate_examples(
        self,
        images: Generator[Tuple[str,IO], None, None],
        masks: Generator[Tuple[str,IO], None, None],
    ):
        for i, (img_info, mask_info) in enumerate(zip(images, masks)):
            img_file_path, img_file_obj = img_info
            mask_file_path, mask_file_obj = mask_info
            yield i, {
                "image": {"path": img_file_path, "bytes": img_file_obj.read()},
                "mask": {"path": mask_file_path, "bytes": mask_file_obj.read()},
            }