# coding=utf-8 # Copyright 2022 The PolyAI and HuggingFace Datasets Authors and the current dataset script contributor. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import os import json import numpy as np import datasets logger = datasets.logging.get_logger(__name__) datasets.Image() """ VTQA Dataset""" _CITATION = """\ @inproceedings{chen2024vtqa, title={VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning}, author={Chen, Kang and Wu, Xiangqian}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, pages={27218--27227}, year={2024} } """ _DESCRIPTION = """\ VTQA is a new dataset containing open-ended questions about image-text pairs. These questions require multimedia entity alignment, multi-step reasoning and open-ended answer generation. """ _HOMEPAGE_URL = "https://visual-text-qa.github.io/" _LICENSE = "The annotations in this dataset belong to the VTQA Consortium and are licensed under a Creative Commons Attribution NonCommercial NoDerivs 4.0 International License" _ALL_CONFIGS = sorted( [ "zh-image", "zh-region", "zh-grid", "en-image", "en-region", "en-grid", "en", "zh", "image", "region", "grid", ] ) _BASE_IMAGE_FEATURES = { "image": datasets.Image(), "region": datasets.Value("string"), "grid": datasets.Value("string"), } _BASE_TEXT_FEATURES = { "raw": { "en": datasets.Value("string"), "zh": datasets.Value("string"), }, "cws": { "en": [datasets.Value("string")], "zh": [datasets.Value("string")], }, } _BASE_ANSWER_FEATURES = { "answer_type": datasets.Value("string"), "answer": { "en": datasets.Value("string"), "zh": datasets.Value("string"), }, } _DATA_URL = "data" class VTQAConfig(datasets.BuilderConfig): """BuilderConfig for VTQA.""" def __init__( self, data_url: str = None, use_cws=False, local_url=None, get_test_split=False, **kwargs ): super(VTQAConfig, self).__init__( version=datasets.Version("1.0.0", ""), description=self.description, **kwargs, ) self.data_url = _DATA_URL if data_url is None else data_url self.use_cws = use_cws self.local_url = local_url self.get_test_split = get_test_split @property def features(self): if self.name == "all": lang, image_type = "all", "all" elif "-" in self.name: lang, image_type = self.name.split("-") elif self.name in ["en", "zh"]: lang, image_type = self.name, "all" elif self.name in ["image", "region", "grid"]: lang, image_type = "all", self.name self.lang, self.image_type = lang, image_type btf = _BASE_TEXT_FEATURES["cws"] if self.use_cws else _BASE_TEXT_FEATURES["raw"] baf = { "answer_type": _BASE_ANSWER_FEATURES["answer_type"], "answer": ( _BASE_ANSWER_FEATURES["answer"] if lang == "all" else _BASE_ANSWER_FEATURES["answer"][lang] ), } dataset_features = datasets.Features( { "question": (btf if lang == "all" else btf[lang]), "question_id": datasets.Value("int64"), "context": (btf if lang == "all" else btf[lang]), "image_id": datasets.Value("int64"), "image_path": ( _BASE_IMAGE_FEATURES if image_type == "all" else _BASE_IMAGE_FEATURES[image_type] ), "answers": [baf], } ) return dataset_features def _build_config(name, use_cws=False, local_url=None): return VTQAConfig( name=name, data_url=_DATA_URL, use_cws=use_cws, local_url=local_url, ) class VTQA(datasets.GeneratorBasedBuilder): BUILDER_CONFIG_CLASS = VTQAConfig DEFAULT_WRITER_BATCH_SIZE = 1000 BUILDER_CONFIGS = [_build_config(name) for name in _ALL_CONFIGS + ["all"]] def _info(self): return datasets.DatasetInfo( description=_DESCRIPTION, features=self.config.features, homepage=_HOMEPAGE_URL, license=_LICENSE, citation=_CITATION, ) def _split_generators(self, dl_manager): lang, image_type = self.config.lang, self.config.image_type def _get_url(file_name): if self.config.local_url is None: return dl_manager.download_and_extract( os.path.join(self.config.data_url, f"{file_name}.zip") ) else: return os.path.join(self.config.local_url, f"{file_name}") annotation_dir = _get_url("annotations") image_dir, region_dir, grid_dir = None, None, None if image_type in ["image", "all"]: image_dir = _get_url("image") if image_type in ["region", "all"]: region_dir = _get_url("region") if image_type in ["grid", "all"]: grid_dir = _get_url("grid") if self.config.use_cws: cws_supp_dir = _get_url("cws_supp") self.cws_supp_dir = cws_supp_dir datasets_split = [ datasets.SplitGenerator( name=datasets.Split.TRAIN, gen_kwargs={ "filepath": os.path.join(annotation_dir, "train.json"), "image_dir": ( os.path.join(image_dir, "train") if image_dir else None ), "region_dir": ( os.path.join(region_dir, "train") if region_dir else None ), "grid_dir": os.path.join(grid_dir, "train") if grid_dir else None, }, ), datasets.SplitGenerator( name=datasets.Split.VALIDATION, gen_kwargs={ "filepath": os.path.join(annotation_dir, "val.json"), "image_dir": ( os.path.join(image_dir, "val") if image_dir else None ), "region_dir": ( os.path.join(region_dir, "val") if region_dir else None ), "grid_dir": os.path.join(grid_dir, "val") if grid_dir else None, }, ), datasets.SplitGenerator( name=datasets.Split("test_dev"), gen_kwargs={ "filepath": os.path.join(annotation_dir, "test_dev.json"), "image_dir": ( os.path.join(image_dir, "test_dev") if image_dir else None ), "region_dir": ( os.path.join(region_dir, "test_dev") if region_dir else None ), "grid_dir": ( os.path.join(grid_dir, "test_dev") if grid_dir else None ), "labeled": False, }, ), ] if self.config.get_test_split: return datasets_split + [ datasets.SplitGenerator( name=datasets.Split.TEST, gen_kwargs={ "filepath": os.path.join(annotation_dir, "test.json"), "image_dir": ( os.path.join(image_dir, "test") if image_dir else None ), "region_dir": ( os.path.join(region_dir, "test") if region_dir else None ), "grid_dir": ( os.path.join(grid_dir, "test") if grid_dir else None ), "labeled": False, }, ) ] else: return datasets_split def _generate_examples( self, filepath, image_dir=None, region_dir=None, grid_dir=None, labeled=True ): lang, image_type = self.config.lang, self.config.image_type use_cws = "cws" if self.config.use_cws else "raw" """Yields examples as (key, example) tuples.""" with open(filepath, encoding="utf-8") as f: vtqa = json.load(f) for id_, d in enumerate(vtqa): text_dict = { "question": ( d["question"][use_cws] if lang == "all" else d["question"][use_cws][lang] ), "context": ( d["context"][use_cws] if lang == "all" else d["context"][use_cws][lang] ), } image_dict = {} if image_dir is not None: image_dict["image"] = os.path.join( image_dir, d["image_name"]["image"] ) if region_dir is not None: image_dict["region"] = os.path.join( region_dir, d["image_name"]["region"] ) if grid_dir is not None: image_dict["grid"] = os.path.join(grid_dir, d["image_name"]["grid"]) if labeled: yield id_, { "question_id": d["question_id"], "image_id": d["image_id"], "answers": [ { "answer_type": a["answer_type"], "answer": ( a["answer"] if lang == "all" else a["answer"][lang] ), } for a in d["answers"] ], **text_dict, "image_path": ( image_dict if image_type == "all" else image_dict[image_type] ), } else: yield id_, { "question_id": d["question_id"], "image_id": d["image_id"], "answers": None, **text_dict, "image_path": ( image_dict if image_type == "all" else image_dict[image_type] ), }