Ammar-alhaj-ali's picture
Update app.py
0b3d03a
raw
history blame
4.62 kB
import os
os.system('pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu')
os.system('pip install -q git+https://github.com/huggingface/transformers.git')
os.system('pip install 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("Ammar-alhaj-ali/LayoutLMv3-Fine-Tuning-FUNSD", apply_ocr=True)
model = AutoModelForTokenClassification.from_pretrained("Ammar-alhaj-ali/LayoutLMv3-Fine-Tuning-FUNSD")
# load image example
#dataset = load_dataset("nielsr/funsd-layoutlmv3", split="test")
#Image.open(dataset[2]["image_path"]).convert("RGB").save("img1.png")
#Image.open(dataset[1]["image_path"]).convert("RGB").save("img2.png")
#Image.open(dataset[0]["image_path"]).convert("RGB").save("img3.png")
# define id2label, label2color
labels = ['O', 'B-HEADER', 'I-HEADER', 'B-QUESTION', 'I-QUESTION', 'B-ANSWER', 'I-ANSWER']
id2label = {v: k for v, k in enumerate(labels)}
label2color = {
"B-HEADER": 'red',
"I-HEADER": 'green',
"B-QUESTION": 'orange',
"I-QUESTION": "blue",
"B-ANSWER": 'gray',
"I-ANSWERE": 'violet',
"O": 'orange'
}
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)
return image
title = "Extraction d'informations de factures en utilisant le modèle LayoutLMv3"
description = "J'utilise LayoutLMv3 de Microsoft formé sur un ensemble de données de factures pour prédire le nom de l'émetteur de factures, l'adresse de l'émetteur de factures, le code postal de l'émetteur de factures, la date d'échéance, la TPS, la date de facturation, le numéro de facture, le sous-total et le total. Pour l'utiliser, il suffit de télécharger une image ou d'utiliser l'exemple d'image ci-dessous. Les résultats seront affichés en quelques secondes."
article="<b>References</b><br>[1] Y. Xu et al., “LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking.” 2022. <a href='https://arxiv.org/abs/2204.08387'>Paper Link</a><br>[2] <a href='https://github.com/NielsRogge/Transformers-Tutorials/tree/master/LayoutLMv3'>LayoutLMv3 training and inference</a>"
examples =[['img1.png'],['img2.png'],['img3.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)