handler.py

import os
import numpy as np
from typing import  Dict, List, Any
from PIL import Image, ImageDraw, ImageFont
from transformers import LayoutLMv2Processor, LayoutLMv2ForTokenClassification
from transformers import pipeline, AutoTokenizer

os.system('apt-get install gcc -y')
os.system('pip3 install pycocotools')
os.system('pip3 install -q detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html')

import detectron2

print(f"DETECTRON2 {detectron2.__version__}")

class EndpointHandler():
    def __init__(self, path=""):
        # load the processor and model

        self.processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
        self.model = LayoutLMv2ForTokenClassification.from_pretrained("nielsr/layoutlmv2-finetuned-funsd")   
        self.id2label = {
                        0: 'O',
                        1: 'B-HEADER',
                        2: 'I-HEADER',
                        3: 'B-QUESTION',
                        4: 'I-QUESTION',
                        5: 'B-ANSWER',
                        6: 'I-ANSWER'
            }


    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
        """
        Args:
            data (:obj:):
                includes the input data and the parameters for the inference.
        Return:
            A :obj:`list`:. The object returned should be a list of one list like [[{"label": 0.9939950108528137}]] containing :
                - "label": A string representing what the label/class is. There can be multiple labels.
                - "score": A score between 0 and 1 describing how confident the model is for this label/class.
        """

        def unnormalize_box(bbox, width, height):
            return [
                width * (bbox[0] / 1000),
                height * (bbox[1] / 1000),
                width * (bbox[2] / 1000),
                height * (bbox[3] / 1000),
            ]

        image = data.pop("inputs", data)
        # encode
        encoding = self.processor(image, truncation=True, return_offsets_mapping=True, return_tensors="pt")
        offset_mapping = encoding.pop('offset_mapping')

        # forward pass
        outputs = self.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
        width, height = image.size

        true_predictions = [self.id2label[prediction]  for prediction in predictions]
        true_boxes = [unnormalize_box(box, width, height) for box in token_boxes]
        is_subword = np.array(offset_mapping.squeeze().tolist())[:,0] != 0

        # postprocess the prediction
        return {"labels": true_predictions, "boxes": true_boxes, "is_subword": is_subword}