import os import datasets from huggingface_hub import HfApi from datasets import DownloadManager, DatasetInfo from datasets.data_files import DataFilesDict _EXTENSION = [".png", ".jpg", ".jpeg"] _NAME = "animelover/danbooru2022" _REVISION = "main" class DanbooruDataset(datasets.GeneratorBasedBuilder): BUILDER_CONFIGS = [ # add number before name for sorting datasets.BuilderConfig( name="0-sfw", description="sfw subset", ), datasets.BuilderConfig( name="1-full", description="full dataset", ), datasets.BuilderConfig( name="2-tags", description="only tags of dataset", ), ] def _info(self) -> DatasetInfo: if self.config.name == "2-tags": features = { "tags": datasets.Value("string"), "post_id": datasets.Value("int64") } else: features = { "image": datasets.Image(), "tags": datasets.Value("string"), "post_id": datasets.Value("int64") } return datasets.DatasetInfo( description=self.config.description, features=datasets.Features(features), supervised_keys=None, citation="", ) def _split_generators(self, dl_manager: DownloadManager): hfh_dataset_info = HfApi().dataset_info(_NAME, revision=_REVISION, timeout=100.0) data_files = DataFilesDict.from_hf_repo( {datasets.Split.TRAIN: ["**"]}, dataset_info=hfh_dataset_info, allowed_extensions=["zip", ".zip"], ) gs = [] for split, files in data_files.items(): downloaded_files = dl_manager.download_and_extract(files) gs.append(datasets.SplitGenerator(name=split, gen_kwargs={"filepath": downloaded_files})) return gs def _generate_examples(self, filepath): for path in filepath: all_fnames = {os.path.relpath(os.path.join(root, fname), start=path) for root, _dirs, files in os.walk(path) for fname in files} image_fnames = sorted([fname for fname in all_fnames if os.path.splitext(fname)[1].lower() in _EXTENSION], reverse=True) for image_fname in image_fnames: image_path = os.path.join(path, image_fname) tags_path = os.path.join(path, os.path.splitext(image_fname)[0] + ".txt") with open(tags_path, "r", encoding="utf-8") as f: tags = f.read() if self.config.name == "0-sfw" and any(tag.strip() in nsfw_tags for tag in tags.split(",")): continue post_id = int(os.path.splitext(os.path.basename(image_fname))[0]) if self.config.name == "2-tags": yield image_fname, {"tags": tags, "post_id": post_id} else: yield image_fname, {"image": image_path, "tags": tags, "post_id": post_id} nsfw_tags = ["nude", "completely nude", "topless", "bottomless", "sex", "oral", "fellatio gesture", "tentacle sex", "nipples", "pussy", "vaginal", "pubic hair", "anus", "ass focus", "penis", "cum", "condom", "sex toy"]