food_vision_199_classes.py

"""
Loading script for the Food Vision 199 classes dataset.

See the template: https://github.com/huggingface/datasets/blob/main/templates/new_dataset_script.py
See the example for Food101: https://huggingface.co/datasets/food101/blob/main/food101.py 
See another example: https://huggingface.co/datasets/davanstrien/encyclopedia_britannica/blob/main/encyclopedia_britannica.py
"""

import datasets
import os
import pandas as pd

from datasets.tasks import ImageClassification

_HOMEPAGE = "https://www.nutrify.app"
_LICENSE = "TODO"
_CITATION = "TODO"
_DESCRIPTION = "Images of 199 food classes from the Nutrify app."

# Read lines of class_names.txt
with open("class_names.txt", "r") as f:
    _NAMES = f.read().splitlines()

class Food199(datasets.GeneratorBasedBuilder):
    """Food199 Images dataset"""

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "image": datasets.Image(),
                    "label": datasets.ClassLabel(names=_NAMES)
                }
            ),
            supervised_keys=("image", "label"),
            homepage=_HOMEPAGE,
            citation=_CITATION,
            license=_LICENSE,
            task_templates=[ImageClassification(image_column="image", label_column="label")],
        )

    def _split_generators(self, dl_manager):
        """
        This function returns the logic to split the dataset into different splits as well as labels.
        """
        csv = dl_manager.download("annotations.csv")
        df = pd.read_csv(csv)
        df_train_annotations = df[df["split"] == "train"].to_dict(orient="records")
        df_test_annotations = df[df["split"] == "test"].to_dict(orient="records")

        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN,
                gen_kwargs={
                    "annotations": df_train_annotations,
                }),
            datasets.SplitGenerator(name=datasets.Split.TEST,
                gen_kwargs={
                    "annotations": df_test_annotations,
                })]


    def _generate_examples(self, annotations):
        """
        This function takes in the kwargs from the _split_generators method and can then yield information from them.
        """
        for id_, row in enumerate(annotations):
            row["image"] = row.pop("filename")
            row["label"] = row.pop("label")
            yield id_, row