dogs-video-object-tracking-dataset.py

import os
from xml.etree import ElementTree as ET

import datasets

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

_DESCRIPTION = """\
The dataset contains frames extracted from videos with dogs on the streets.
Each frame is accompanied by **bounding box** that specifically **tracks the dog**
in the image.
The dataset provides a valuable resource for advancing computer vision tasks,
enabling the development of more accurate and effective solutions for monitoring and
understanding dog behavior in urban settings.
"""
_NAME = "dogs-video-object-tracking-dataset"

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

_LICENSE = ""

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

_LABELS = ["dog"]


class DogsVideoObjectTrackingDataset(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="video_01", data_dir=f"{_DATA}video_01.zip"),
        datasets.BuilderConfig(name="video_02", data_dir=f"{_DATA}video_02.zip"),
        datasets.BuilderConfig(name="video_03", data_dir=f"{_DATA}video_03.zip"),
    ]

    DEFAULT_CONFIG_NAME = "video_01"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "id": datasets.Value("int32"),
                    "name": datasets.Value("string"),
                    "image": datasets.Image(),
                    "mask": datasets.Image(),
                    "shapes": datasets.Sequence(
                        {
                            "track_id": datasets.Value("uint32"),
                            "label": datasets.ClassLabel(
                                num_classes=len(_LABELS),
                                names=_LABELS,
                            ),
                            "type": datasets.Value("string"),
                            "points": datasets.Sequence(
                                datasets.Sequence(
                                    datasets.Value("float"),
                                ),
                            ),
                            "rotation": datasets.Value("float"),
                            "occluded": datasets.Value("uint8"),
                            "attributes": datasets.Sequence(
                                {
                                    "name": datasets.Value("string"),
                                    "text": datasets.Value("string"),
                                }
                            ),
                        }
                    ),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        data = dl_manager.download_and_extract(self.config.data_dir)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "data": data,
                },
            ),
        ]

    @staticmethod
    def extract_shapes_from_tracks(
        root: ET.Element, file: str, index: int
    ) -> ET.Element:
        img = ET.Element("image")
        img.set("name", file)
        img.set("id", str(index))
        for track in root.iter("track"):
            shape = track.find(f".//*[@frame='{index}']")
            if not (shape is None):
                shape.set("label", track.get("label"))
                shape.set("track_id", track.get("id"))
                img.append(shape)

        return img

    @staticmethod
    def parse_shape(shape: ET.Element) -> dict:
        label = shape.get("label")
        track_id = shape.get("track_id")
        shape_type = shape.tag
        rotation = shape.get("rotation", 0.0)
        occluded = shape.get("occluded", 0)

        points = None

        if shape_type == "points":
            points = tuple(map(float, shape.get("points").split(",")))

        elif shape_type == "box":
            points = [
                (float(shape.get("xtl")), float(shape.get("ytl"))),
                (float(shape.get("xbr")), float(shape.get("ybr"))),
            ]

        elif shape_type == "polygon":
            points = [
                tuple(map(float, point.split(",")))
                for point in shape.get("points").split(";")
            ]

        attributes = []

        for attr in shape:
            attr_name = attr.get("name")
            attr_text = attr.text
            attributes.append({"name": attr_name, "text": attr_text})

        shape_data = {
            "label": label,
            "track_id": track_id,
            "type": shape_type,
            "points": points,
            "rotation": rotation,
            "occluded": occluded,
            "attributes": attributes,
        }

        return shape_data

    def _generate_examples(self, data):
        tree = ET.parse(os.path.join(data, "annotations.xml"))
        root = tree.getroot()

        for idx, file in enumerate(sorted(os.listdir(os.path.join(data, "images")))):
            img = self.extract_shapes_from_tracks(root, file, idx)

            image_id = img.get("id")
            name = img.get("name")
            shapes = [self.parse_shape(shape) for shape in img]
            print(shapes)

            yield idx, {
                "id": image_id,
                "name": name,
                "image": os.path.join(data, "images", file),
                "mask": os.path.join(data, "masks", file),
                "shapes": shapes,
            }