fish-tracking-dataset.py

import datasets
import pandas as pd

_CITATION = """\
@InProceedings{huggingface:dataset,
title = {fish-tracking-dataset},
author = {TrainingDataPro},
year = {2023}
}
"""

_DESCRIPTION = """\
The collection of video frames, capturing various types of fish swimming in the water.
The dataset includes fish of different colors, sizes and with different swimming speeds.
"""
_NAME = "fish-tracking-dataset"

_HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"

_LICENSE = ""

_DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"


class FishTrackingDataset(datasets.GeneratorBasedBuilder):
    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("int32"),
                    "image": datasets.Image(),
                    "mask": datasets.Image(),
                    "bboxes": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        images = dl_manager.download(f"{_DATA}images.tar.gz")
        masks = dl_manager.download(f"{_DATA}boxes.tar.gz")
        annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
        images = dl_manager.iter_archive(images)
        masks = dl_manager.iter_archive(masks)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "images": images,
                    "masks": masks,
                    "annotations": annotations,
                },
            ),
        ]

    def _generate_examples(self, images, masks, annotations):
        annotations_df = pd.read_csv(annotations)

        for idx, ((image_path, image), (mask_path, mask)) in enumerate(
            zip(images, masks)
        ):
            yield idx, {
                "id": annotations_df["image_id"].iloc[idx],
                "image": {"path": image_path, "bytes": image.read()},
                "mask": {"path": mask_path, "bytes": mask.read()},
                "bboxes": annotations_df["annotations"].iloc[idx],
            }