app.py

import os
os.system('pip install git+https://github.com/huggingface/transformers.git --upgrade')
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')

import gradio as gr
import numpy as np
from transformers import LayoutLMv2FeatureExtractor, LayoutLMv2Tokenizer, LayoutLMV2ForTokenClassification
from datasets import load_dataset
from PIL import Image, ImageDraw, ImageFont

ds = load_dataset("hf-internal-testing/fixtures_docvqa", split="test")

image = Image.open(ds[0]["file"]).convert("RGB")
image.save("document.png")

feature_extractor = LayoutLMv2FeatureExtractor.from_pretrained("microsoft/layoutlmv2-base-uncased")
tokenizer = LayoutLMv2Tokenizer.from_pretrained("microsoft/layoutlmv2-base-uncased")
model = LayoutLMv2ForTokenClassification.from_pretrained("nielsr/layoutlmv2-finetuned-funsd")

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):
    label = label[2:]
    if not label:
      return 'other'
    return label

def process_image(image):
    width, height = image.size
    
    # get words, boxes
    encoding_feature_extractor = feature_extractor(image, return_tensors="pt")
    words, boxes = encoding_feature_extractor.words, encoding_feature_extractor.boxes

    # encode
    encoding = tokenizer(words, boxes=boxes, return_offsets_mapping=True, return_tensors="pt")
    offset_mapping = encoding.pop('offset_mapping')
    encoding["image"] = encoding_feature_extractor.pixel_values

    # 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).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 = "Interactive demo: LayoutLMv2"
description = "Demo for Microsoft's LayoutLMv2, a Transformer for state-of-the-art document image understanding tasks. To use it, simply upload an image or use the example image below. Results will show up in a few seconds."
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2012.14740'>LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding</a> | <a href='https://github.com/microsoft/unilm'>Github Repo</a></p>"
examples =[['document.png']]

iface = gr.Interface(fn=process_image, 
                     inputs=gr.inputs.Image(shape=(480, 480), type="pil"), 
                     outputs=gr.outputs.Image(type='pil', label=f'annotated image'),
                     title=title,
                     description=description,
                     article=article,
                     examples=examples)
iface.launch()