# -*- coding: utf-8 -*- """DocAI_DeploymentGradio.ipynb Automatically generated by Colaboratory. Original file is located at https://colab.research.google.com/drive/1USSEj7nHh2n2hUhTJTC0Iwhj6mSR7-mD """ import os os.system('pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu') os.system('pip install pyyaml==5.1') os.system('pip install -q git+https://github.com/huggingface/transformers.git') os.system('pip install -q datasets seqeval') os.system('pip install torch==1.8.0+cu101 torchvision==0.9.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html') os.system('pip install -q detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/torch1.8/index.html') os.system('pip install -q pytesseract') #!pip install gradio !pip install -q git+https://github.com/huggingface/transformers.git !pip install h5py !pip install -q datasets seqeval import gradio as gr import numpy as np import tensorflow as tf import torch import json from datasets.features import ClassLabel from transformers import AutoProcessor from datasets import Features, Sequence, ClassLabel, Value, Array2D, Array3D from datasets import load_dataset # this dataset uses the new Image feature :) from transformers import LayoutLMv3Processor,LayoutLMv3ForTokenClassification, AutoProcessor ,AutoModelForTokenClassification #import cv2 from PIL import Image, ImageDraw, ImageFont processor = LayoutLMv3Processor.from_pretrained("microsoft/layoutlmv3-base",apply_ocr = True) model = LayoutLMv3ForTokenClassification.from_pretrained("nielsr/layoutlmv3-finetuned-funsd") dataset = load_dataset("nielsr/funsd-layoutlmv3") example = dataset["test"][0] example["image"].save("example1.png") example1 = dataset["test"][1] example1["image"].save("example2.png") example2 = dataset["test"][2] example2["image"].save("example3.png") #example2["image"] labels = dataset["test"].features['ner_tags'].feature.names words, boxes, ner_tags = example["tokens"], example["bboxes"], example["ner_tags"] features = dataset["test"].features column_names = dataset["test"].column_names image_column_name = "image" text_column_name = "tokens" boxes_column_name = "bboxes" label_column_name = "ner_tags" # In the event the labels are not a `Sequence[ClassLabel]`, we will need to go through the dataset to get the # unique labels. def get_label_list(labels): unique_labels = set() for label in labels: unique_labels = unique_labels | set(label) label_list = list(unique_labels) label_list.sort() return label_list if isinstance(features[label_column_name].feature, ClassLabel): label_list = features[label_column_name].feature.names # No need to convert the labels since they are already ints. id2label = {k: v for k,v in enumerate(label_list)} label2id = {v: k for k,v in enumerate(label_list)} else: label_list = get_label_list(dataset["test"][label_column_name]) id2label = {k: v for k,v in enumerate(label_list)} label2id = {v: k for k,v in enumerate(label_list)} num_labels = len(label_list) def get_label_list(labels): unique_labels = set() for label in labels: unique_labels = unique_labels | set(label) label_list = list(unique_labels) label_list.sort() return label_list label2color = {'question':'blue', 'answer':'green', 'header':'orange', 'other':'violet'} def unnormalize_box(bbox, width, height): return [ width * (bbox[0] / 1000), height * (bbox[1] / 1000), width * (bbox[2] / 1000), height * (bbox[3] / 1000), ] # we need to define custom features for `set_format` (used later on) to work properly features = Features({ 'pixel_values': Array3D(dtype="float32", shape=(3, 224, 224)), 'input_ids': Sequence(feature=Value(dtype='int64')), 'attention_mask': Sequence(Value(dtype='int64')), 'bbox': Array2D(dtype="int64", shape=(512, 4)), 'labels': Sequence(feature=Value(dtype='int64')), }) def process_image(image): width, height = image.size # encode encoding = processor(image, truncation=True, return_offsets_mapping=True, return_tensors="pt") offset_mapping = encoding.pop('offset_mapping') # forward pass outputs = model(**encoding) # get predictions predictions = outputs.logits.argmax(-1).squeeze().tolist() token_boxes = encoding.bbox.squeeze().tolist() # only keep non-subword predictions is_subword = np.array(offset_mapping.squeeze().tolist())[:,0] != 0 true_predictions = [id2label[pred] for idx, pred in enumerate(predictions) if not is_subword[idx]] true_boxes = [unnormalize_box(box, width, height) for idx, box in enumerate(token_boxes) if not is_subword[idx]] # draw predictions over the image draw = ImageDraw.Draw(image) font = ImageFont.load_default() def iob_to_label(label): label = label[2:] if not label: return 'other' return label label2color = {'question':'blue', 'answer':'green', 'header':'orange', 'other':'violet'} for prediction, box in zip(true_predictions, true_boxes): predicted_label = iob_to_label(prediction) #.lower() draw.rectangle(box, outline=label2color[predicted_label]) draw.text((box[0]+10, box[1]-10), text=predicted_label, fill=label2color[predicted_label], font=font) return image title = "DocumentAI - Extraction using LayoutLMv3 model" description = "Extraction of Form or Invoice Extraction - We use Microsoft's LayoutLMv3 trained on Invoice Dataset to predict the Biller Name, Biller Address, Biller post_code, Due_date, GST, Invoice_date, Invoice_number, Subtotal and Total. To use it, simply upload an image or use the example image below. Results will show up in a few seconds." article="References
[1] Y. Xu et al., “LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking.” 2022. Paper Link
[2] LayoutLMv3 training and inference" examples =[['example1.png'],['example2.png'],['example3.png']] css = """.output_image, .input_image {height: 600px !important}""" iface = gr.Interface(fn=process_image, inputs=gr.inputs.Image(type="pil"), outputs=gr.outputs.Image(type="pil", label="annotated image"), title=title, description=description, article=article, examples=examples, css=css, analytics_enabled = True, enable_queue=True ) #iface.launch(inline=False, share=False, debug=False) iface.launch(inline=False)