GraySpectrogram.py

import datasets
from huggingface_hub import HfApi
from datasets import DownloadManager, DatasetInfo
from datasets.data_files import DataFilesDict
import os
import json
from os.path import dirname, basename
from pathlib import Path


# ここに設定を記入
_NAME = "mb23/GraySpectrogram"
_EXTENSION = [".png"]
_REVISION = "main"

# _HOMEPAGE = "https://github.com/fastai/imagenette"
# プログラムを置く場所が決まったら、ここにホームページURLつける
_HOMEPAGE = "https://huggingface.co/datasets/mickylan2367/spectrogram_musicCaps"

_DESCRIPTION = f"""\
{_NAME} Datasets including spectrogram.png file from Google MusicCaps Datasets!
Using for Project Learning...
"""

# 参考になりそうなURL集
# https://huggingface.co/docs/datasets/v1.1.1/_modules/datasets/utils/download_manager.html
# https://huggingface.co/docs/datasets/package_reference/builder_classes
# https://huggingface.co/datasets/animelover/danbooru2022/blob/main/danbooru2022.py
# https://huggingface.co/datasets/food101/blob/main/food101.py
# https://huggingface.co/docs/datasets/about_dataset_load
# https://huggingface.co/datasets/frgfm/imagenette/blob/main/imagenette.py
# https://huggingface.co/docs/datasets/v1.2.1/add_dataset.html
# DatasetInfo : https://huggingface.co/docs/datasets/package_reference/main_classes



class GraySpectrogram2(datasets.GeneratorBasedBuilder):

    # データのサブセットはここで用意
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="train",
            description=_DESCRIPTION,
        )
    ]

    def _info(self) -> DatasetInfo:
      return datasets.DatasetInfo(
          description = self.config.description,
          features=datasets.Features(
              {
                  "image": datasets.Image(),
                  "caption": datasets.Value("string"),
                  "data_idx": datasets.Value("int32"),
                  "number" : datasets.Value("int32"),
                  "label" : datasets.ClassLabel(
                        names=[
                            "blues",
                            "classical",
                            "country",
                            "disco",
                            "hiphop",
                            "metal",
                            "pop",
                            "reggae",
                            "rock",
                            "jazz"
                        ]
                  )
              }
          ),
          supervised_keys=("image", "caption"),
          homepage=_HOMEPAGE,
          citation= "",
          # license=_LICENSE,
          # task_templates=[ImageClassification(image_column="image", label_column="label")],
      )

    def _split_generators(self, dl_manager: DownloadManager):
        # huggingfaceのディレクトリからデータを取ってくる
        hfh_dataset_info = HfApi().dataset_info(_NAME, revision=_REVISION, timeout=100.0)

        # メタデータであるjsonlファイルのURLを取得
        metadata_urls = DataFilesDict.from_hf_repo(
            {datasets.Split.TRAIN: ["**"]},
            dataset_info=hfh_dataset_info,
            allowed_extensions=["jsonl", ".jsonl"],
        )

        # 辞書型にしてURLを格納し直す <- 正しい辞書型にできていたら、必要ないかも
        metadata_paths = dict()
        for path in metadata_urls["train"]:
            dname = dirname(path)
            folder = basename(Path(dname))
            # fname = basename(path)
            metadata_paths[folder] = path

        # 画像データは**.zipのURLをDict型として取得？
        data_urls = DataFilesDict.from_hf_repo(
            {datasets.Split.TRAIN: ["**"]},
            dataset_info=hfh_dataset_info,
            allowed_extensions=["zip", ".zip"],
        )

        data_url = dict()
        train_data_url = list()
        test_data_url = list()
        for path in data_urls["train"]:
            dname = dirname(path)
            folder = basename(Path(dname))
            # 辞書型
            if folder=="train":
                train_data_url.append(path)
            if folder == "test":
                test_data_url.append(path)

        data_url["train"] = train_data_url
        data_url["test"] = test_data_url

        # iteration
        iter_archive = dict()
        for split, files in data_url.items():
            file_name_obj = list()
            for file_ in files:
                downloaded_files_path = dl_manager.download(file_)
                for file_obj in dl_manager.iter_archive(downloaded_files_path):
                    # print(file_obj)
                    if file_obj[0].startswith('content/'):
                        fname = basename(file_obj[0])
                        file_obj = (fname, file_obj[1])
                        file_name_obj.append(file_obj)
                        # print(file_obj)
                    else:
                        file_name_obj.append(file_obj)
                        # print(file_obj)
            iter_archive[split] = file_name_obj
        
        gs = []
        for split, files in iter_archive.items():
            '''
            split : "train" or "test" or "val"
            files : zip files
            '''
            # リポジトリからダウンロードしてとりあえずキャッシュしたURLリストを取得
            metadata_path = dl_manager.download_and_extract(metadata_paths[split])
            # 元のコードではzipファイルの中身を"filepath"としてそのまま_generate_exampleに引き渡している？
            gs.append(
                datasets.SplitGenerator(
                    name = split,
                    gen_kwargs = {
                        "images" : iter(iter_archive[split]),
                        "metadata_path": metadata_path # メタデータパスを渡す
                    }
                )
            )
        return gs

    
    def _generate_examples(self, images, metadata_path):
        """Generate images and captions for splits."""
        # with open(metadata_path, encoding="utf-8") as f:
        #     files_to_keep = set(f.read().split("\n"))
        file_list = list()
        caption_list = list()
        dataIDX_list = list()
        num_list = list()
        label_list = list()

        with open(metadata_path) as fin:
            for line in fin:
                data =  json.loads(line)
                file_list.append(data["file_name"])
                caption_list.append(data["caption"])
                dataIDX_list.append(data["data_idx"])
                num_list.append(data["number"])
                label_list.append(data["label"])

        for idx, (file_path, file_obj) in enumerate(images):
            yield file_path, {
                "image": {
                    "path": file_path, 
                    "bytes": file_obj.read()
                },
                "caption" : caption_list[idx],
                "data_idx" : dataIDX_list[idx],
                "number" : num_list[idx],
                "label": label_list[idx]
            }