LayoutLM fine-tuned on FUNSD for Document/Forms token classification

Usage (WIP)

import torch
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import pytesseract
from transformers import LayoutLMForTokenClassification, LayoutLMTokenizer


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

tokenizer = LayoutLMTokenizer.from_pretrained("mrm8488/layoutlm-finetuned-funsd")
model = LayoutLMForTokenClassification.from_pretrained("mrm8488/layoutlm-finetuned-funsd", num_labels=13)
model.to(device)


image = Image.open("/83443897.png")
image = image.convert("RGB")

# Display the image


# Run Tesseract (OCR) on the image

width, height = image.size
w_scale = 1000/width
h_scale = 1000/height

ocr_df = pytesseract.image_to_data(image, output_type='data.frame') \\n            
ocr_df = ocr_df.dropna() \\n               .assign(left_scaled = ocr_df.left*w_scale,
                       width_scaled = ocr_df.width*w_scale,
                       top_scaled = ocr_df.top*h_scale,
                       height_scaled = ocr_df.height*h_scale,
                       right_scaled = lambda x: x.left_scaled + x.width_scaled,
                       bottom_scaled = lambda x: x.top_scaled + x.height_scaled)

float_cols = ocr_df.select_dtypes('float').columns
ocr_df[float_cols] = ocr_df[float_cols].round(0).astype(int)
ocr_df = ocr_df.replace(r'^\s*$', np.nan, regex=True)
ocr_df = ocr_df.dropna().reset_index(drop=True)
ocr_df[:20]

# create a list of words, actual bounding boxes, and normalized boxes

words = list(ocr_df.text)
coordinates = ocr_df[['left', 'top', 'width', 'height']]
actual_boxes = []
for idx, row in coordinates.iterrows():
  x, y, w, h = tuple(row) # the row comes in (left, top, width, height) format
  actual_box = [x, y, x+w, y+h] # we turn it into (left, top, left+widght, top+height) to get the actual box 
  actual_boxes.append(actual_box)

def normalize_box(box, width, height):
    return [
        int(1000 * (box[0] / width)),
        int(1000 * (box[1] / height)),
        int(1000 * (box[2] / width)),
        int(1000 * (box[3] / height)),
    ]

boxes = []
for box in actual_boxes:
  boxes.append(normalize_box(box, width, height))
 
# Display boxes

def convert_example_to_features(image, words, boxes, actual_boxes, tokenizer, args, cls_token_box=[0, 0, 0, 0],
                                 sep_token_box=[1000, 1000, 1000, 1000],
                                 pad_token_box=[0, 0, 0, 0]):
      width, height = image.size

      tokens = []
      token_boxes = []
      actual_bboxes = [] # we use an extra b because actual_boxes is already used
      token_actual_boxes = []
      for word, box, actual_bbox in zip(words, boxes, actual_boxes):
          word_tokens = tokenizer.tokenize(word)
          tokens.extend(word_tokens)
          token_boxes.extend([box] * len(word_tokens))
          actual_bboxes.extend([actual_bbox] * len(word_tokens))
          token_actual_boxes.extend([actual_bbox] * len(word_tokens))

      # Truncation: account for [CLS] and [SEP] with "- 2". 
      special_tokens_count = 2 
      if len(tokens) > args.max_seq_length - special_tokens_count:
          tokens = tokens[: (args.max_seq_length - special_tokens_count)]
          token_boxes = token_boxes[: (args.max_seq_length - special_tokens_count)]
          actual_bboxes = actual_bboxes[: (args.max_seq_length - special_tokens_count)]
          token_actual_boxes = token_actual_boxes[: (args.max_seq_length - special_tokens_count)]

      # add [SEP] token, with corresponding token boxes and actual boxes
      tokens += [tokenizer.sep_token]
      token_boxes += [sep_token_box]
      actual_bboxes += [[0, 0, width, height]]
      token_actual_boxes += [[0, 0, width, height]]
      
      segment_ids = [0] * len(tokens)

      # next: [CLS] token
      tokens = [tokenizer.cls_token] + tokens
      token_boxes = [cls_token_box] + token_boxes
      actual_bboxes = [[0, 0, width, height]] + actual_bboxes
      token_actual_boxes = [[0, 0, width, height]] + token_actual_boxes
      segment_ids = [1] + segment_ids

      input_ids = tokenizer.convert_tokens_to_ids(tokens)

      # The mask has 1 for real tokens and 0 for padding tokens. Only real
      # tokens are attended to.
      input_mask = [1] * len(input_ids)

      # Zero-pad up to the sequence length.
      padding_length = args.max_seq_length - len(input_ids)
      input_ids += [tokenizer.pad_token_id] * padding_length
      input_mask += [0] * padding_length
      segment_ids += [tokenizer.pad_token_id] * padding_length
      token_boxes += [pad_token_box] * padding_length
      token_actual_boxes += [pad_token_box] * padding_length

      assert len(input_ids) == args.max_seq_length
      assert len(input_mask) == args.max_seq_length
      assert len(segment_ids) == args.max_seq_length
      assert len(token_boxes) == args.max_seq_length
      assert len(token_actual_boxes) == args.max_seq_length
      
      return input_ids, input_mask, segment_ids, token_boxes, token_actual_boxes
      
input_ids, input_mask, segment_ids, token_boxes, token_actual_boxes = convert_example_to_features(image=image, words=words, boxes=boxes, actual_boxes=actual_boxes, tokenizer=tokenizer, args=args)

input_ids = torch.tensor(input_ids, device=device).unsqueeze(0)
attention_mask = torch.tensor(input_mask, device=device).unsqueeze(0)
token_type_ids = torch.tensor(segment_ids, device=device).unsqueeze(0)
bbox = torch.tensor(token_boxes, device=device).unsqueeze(0)


outputs = model(input_ids=input_ids, bbox=bbox, attention_mask=attention_mask, token_type_ids=token_type_ids)

token_predictions = outputs.logits.argmax(-1).squeeze().tolist() # the predictions are at the token level

word_level_predictions = [] # let's turn them into word level predictions
final_boxes = []
for id, token_pred, box in zip(input_ids.squeeze().tolist(), token_predictions, token_actual_boxes):
  if (tokenizer.decode([id]).startswith("##")) or (id in [tokenizer.cls_token_id, 
                                                           tokenizer.sep_token_id, 
                                                          tokenizer.pad_token_id]):
    # skip prediction + bounding box

    continue
  else:
    word_level_predictions.append(token_pred)
    final_boxes.append(box)

#print(word_level_predictions)


draw = ImageDraw.Draw(image)

font = ImageFont.load_default()

def iob_to_label(label):
  if label != 'O':
    return label[2:]
  else:
    return "other"

label2color = {'question':'blue', 'answer':'green', 'header':'orange', 'other':'violet'}

for prediction, box in zip(word_level_predictions, final_boxes):
    predicted_label = iob_to_label(label_map[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)

# Display the result (image)