import os os.system('pip install pip --upgrade') os.system('pip install -q git+https://github.com/huggingface/transformers.git') os.system("pip install pyyaml==5.1") # workaround: install old version of pytorch since detectron2 hasn't released packages for pytorch 1.9 (issue: https://github.com/facebookresearch/detectron2/issues/3158) os.system("pip install torch==1.8.0+cu101 torchvision==0.9.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html") # install detectron2 that matches pytorch 1.8 # See https://detectron2.readthedocs.io/tutorials/install.html for instructions os.system( "pip install -q detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/torch1.8/index.html" ) ## install PyTesseract os.system("pip install -q pytesseract") import gradio as gr import numpy as np from transformers import AutoModelForTokenClassification from datasets.features import ClassLabel from transformers import AutoProcessor from datasets import Features, Sequence, ClassLabel, Value, Array2D, Array3D import torch from datasets import load_metric from transformers import LayoutLMv3ForTokenClassification from transformers.data.data_collator import default_data_collator from transformers import AutoModelForTokenClassification from datasets import load_dataset from PIL import Image, ImageDraw, ImageFont processor = AutoProcessor.from_pretrained("jinhybr/OCR-LayoutLMv3-Invoice", apply_ocr=True) model = AutoModelForTokenClassification.from_pretrained("jinhybr/OCR-LayoutLMv3-Invoice") # load image example dataset = load_dataset("jinhybr/WildReceipt", split="test") Image.open(dataset[1]["image_path"]).convert("RGB").save("example1.png") Image.open(dataset[3]["image_path"]).convert("RGB").save("example2.png") Image.open(dataset[25]["image_path"]).convert("RGB").save("example3.png") # define id2label, label2color labels = dataset.features['ner_tags'].feature.names id2label = {v: k for v, k in enumerate(labels)} label2color = { "Date_key": 'red', "Date_value": 'green', "Ignore": 'orange', "Others": 'orange', "Prod_item_key": 'red', "Prod_item_value": 'green', "Prod_price_key": 'red', "Prod_price_value": 'green', "Prod_quantity_key": 'red', "Prod_quantity_value": 'green', "Store_addr_key": 'red', "Store_addr_value": 'green', "Store_name_key": 'red', "Store_name_value": 'green', "Subtotal_key": 'red', "Subtotal_value": 'green', "Tax_key": 'red', "Tax_value": 'green', "Tel_key": 'red', "Tel_value": 'green', "Time_key": 'red', "Time_value": 'green', "Tips_key": 'red', "Tips_value": 'green', "Total_key": 'red', "Total_value": 'blue' } def unnormalize_box(bbox, width, height): return [ width * (bbox[0] / 1000), height * (bbox[1] / 1000), width * (bbox[2] / 1000), height * (bbox[3] / 1000), ] def iob_to_label(label): return label def process_image(image): print(type(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() for prediction, box in zip(true_predictions, true_boxes): predicted_label = iob_to_label(prediction) draw.rectangle(box, outline=label2color[predicted_label]) draw.text((box[0]+10, box[1]-10), text=predicted_label, fill=label2color[predicted_label], font=font) # Print label and value label_text = f"Label: {predicted_label}" value_text = f"Value: {prediction}" print(label_text) print(value_text) print("------") return image title = "OCR Invoice - Information Extraction - LayoutLMv3" description = "Fine-tuned Microsoft's LayoutLMv3 on WildReceipt Dataset to parse Invoice OCR document. 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
[3] Hongbin Sun, Zhanghui Kuang, Xiaoyu Yue, Chenhao Lin, and Wayne Zhang. 2021. Spatial Dual-Modality Graph Reasoning for Key Information Extraction. arXiv. DOI:https://doi.org/10.48550/ARXIV.2103.14470 Paper Link" examples =[['example1.png'],['example2.png'],['example3.png'],['inv2.jpg']] 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=True)