# Modified from https://huggingface.co/datasets/rogerdehe/xfund/blob/main/xfund.py # -*- coding: utf-8 -*- import json import os from PIL import Image import datasets def load_image(image_path): image = Image.open(image_path).convert("RGB") w, h = image.size return image, (w, h) def normalize_bbox(bbox, size): width, height = size def clip(min_num, num, max_num): return min(max(num, min_num), max_num) x0, y0, x1, y1 = bbox x0 = clip(0, int((x0 / width) * 1000), 1000) y0 = clip(0, int((y0 / height) * 1000), 1000) x1 = clip(0, int((x1 / width) * 1000), 1000) y1 = clip(0, int((y1 / height) * 1000), 1000) assert x1 >= x0 assert y1 >= y0 return [x0, y0, x1, y1] logger = datasets.logging.get_logger(__name__) _CITATION = """\ @inproceedings{xu-etal-2022-xfund, title = "{XFUND}: A Benchmark Dataset for Multilingual Visually Rich Form Understanding", author = "Xu, Yiheng and Lv, Tengchao and Cui, Lei and Wang, Guoxin and Lu, Yijuan and Florencio, Dinei and Zhang, Cha and Wei, Furu", booktitle = "Findings of the Association for Computational Linguistics: ACL 2022", month = may, year = "2022", address = "Dublin, Ireland", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2022.findings-acl.253", doi = "10.18653/v1/2022.findings-acl.253", pages = "3214--3224", abstract = "Multimodal pre-training with text, layout, and image has achieved SOTA performance for visually rich document understanding tasks recently, which demonstrates the great potential for joint learning across different modalities. However, the existed research work has focused only on the English domain while neglecting the importance of multilingual generalization. In this paper, we introduce a human-annotated multilingual form understanding benchmark dataset named XFUND, which includes form understanding samples in 7 languages (Chinese, Japanese, Spanish, French, Italian, German, Portuguese). Meanwhile, we present LayoutXLM, a multimodal pre-trained model for multilingual document understanding, which aims to bridge the language barriers for visually rich document understanding. Experimental results show that the LayoutXLM model has significantly outperformed the existing SOTA cross-lingual pre-trained models on the XFUND dataset. The XFUND dataset and the pre-trained LayoutXLM model have been publicly available at https://aka.ms/layoutxlm.", } """ _DESCRIPTION = """\ https://github.com/doc-analysis/XFUND """ _LANG = ["de", "es", "fr", "it", "ja", "pt", "zh"] _URL = "https://github.com/doc-analysis/XFUND/releases/download/v1.0" class XFund(datasets.GeneratorBasedBuilder): """XFund dataset.""" BUILDER_CONFIGS = [ datasets.BuilderConfig(name=f"{lang}", version=datasets.Version("1.0.0"), description=f"XFUND {lang} dataset") for lang in _LANG ] def _info(self): return datasets.DatasetInfo( description=_DESCRIPTION, features=datasets.Features( { "id": datasets.Value("string"), "tokens": datasets.Sequence(datasets.Value("string")), "bboxes": datasets.Sequence(datasets.Sequence(datasets.Value("int64"))), "tags": datasets.Sequence( datasets.features.ClassLabel( # names=["O", "HEADER", "QUESTION", "ANSWER"] names=["O", "B-HEADER", "I-HEADER", "B-QUESTION", "I-QUESTION", "B-ANSWER", "I-ANSWER"] ) ), "image": datasets.features.Image(), } ), supervised_keys=None, homepage="https://github.com/doc-analysis/XFUND", citation=_CITATION, ) def _split_generators(self, dl_manager): """Returns SplitGenerators.""" lang = self.config.name fileinfos = dl_manager.download_and_extract({ "train_image": f"{_URL}/{lang}.train.zip", "train_annotation": f"{_URL}/{lang}.train.json", "valid_image": f"{_URL}/{lang}.val.zip", "valid_annotation": f"{_URL}/{lang}.val.json", }) logger.info(f"file infos: {fileinfos}") return [ datasets.SplitGenerator( name=datasets.Split.TRAIN, gen_kwargs={"image_path": fileinfos['train_image'], "annotation_path": fileinfos["train_annotation"]} ), datasets.SplitGenerator( name=datasets.Split.TEST, gen_kwargs={"image_path": fileinfos["valid_image"], "annotation_path": fileinfos["valid_annotation"]} ), ] # def get_line_bbox(self, bboxs): # x = [bboxs[i][j] for i in range(len(bboxs)) for j in range(0, len(bboxs[i]), 2)] # y = [bboxs[i][j] for i in range(len(bboxs)) for j in range(1, len(bboxs[i]), 2)] # x0, y0, x1, y1 = min(x), min(y), max(x), max(y) # assert x1 >= x0 and y1 >= y0 # bbox = [[x0, y0, x1, y1] for _ in range(len(bboxs))] # return bbox def _generate_examples(self, image_path, annotation_path): logger.info("⏳ Generating examples from = %s %s", image_path, annotation_path) with open(annotation_path) as fi: ann_infos = json.load(fi) document_list = ann_infos["documents"] for guid, doc in enumerate(document_list): tokens, bboxes, tags = list(), list(), list() image_file = os.path.join(image_path, doc["img"]["fname"]) # cannot load image when submit code to huggingface # image, size = load_image(image_file) # assert size[0] == doc["img"]["width"] # assert size[1] == doc["img"]["height"] size = [doc["img"]["width"], doc["img"]["height"]] for item in doc["document"]: # cur_line_bboxes = list() # text, label = item["text"], item["label"] # bbox = normalize_bbox(item["box"], size) # if len(text) == 0: # continue word_box_list_raw = [w['box'] for w in item['words']] word_box_list = [normalize_bbox(w, size) for w in word_box_list_raw] word_text_list = [w['text'] for w in item['words']] if item['label'] == 'other': label_list = ['O'] * len(item['words']) else: label_list = [f'B-{item["label"].upper()}'] + ['I-' + item['label'].upper()] * (len(item['words']) - 1) tokens.extend(word_text_list) bboxes.extend(word_box_list) tags.extend(label_list) yield guid, {"id": doc["id"], "tokens": tokens, "bboxes": bboxes, "tags": tags, "image": Image.open(image_file)}