import streamlit as st from transformers import LayoutLMv3ForTokenClassification, LayoutLMv3Processor from PIL import Image import torch import easyocr from PIL import Image import re # Load the model and processor from Hugging Face model_name = "capitaletech/LayoutLMv3-v1" # Replace with your model repository name model = LayoutLMv3ForTokenClassification.from_pretrained(model_name) processor = LayoutLMv3Processor.from_pretrained(model_name) st.title("LayoutLMv3 Text Extraction") st.write("Upload an image to get text predictions using the fine-tuned LayoutLMv3 model.") uploaded_file = st.file_uploader("Choose an image...", type="png") if uploaded_file is not None: image = Image.open(uploaded_file) st.image(image, caption='Uploaded Image.', use_column_width=True) st.write("") st.write("Classifying...") # Process the image words = uploaded_file["tokens"] boxes = uploaded_file["bboxes"] word_labels = uploaded_file["ner_tags"] encoding = processor(image, words, boxes=boxes, word_labels=word_labels, return_tensors="pt") with torch.no_grad(): outputs = model(**encoding) logits = outputs.logits predictions = logits.argmax(-1).squeeze().cpu.tolist() labels = encoding['labels'].squeeze().tolist() # Set up the EasyOCR reader for multiple languages languages = ["ru", "rs_cyrillic", "be", "bg", "uk", "mn", "en"] reader = easyocr.Reader(languages) # Load the image image_path = example["img_path"] image = Image.open(image_path) # Perform text detection ocr_results = reader.readtext(image_path, detail=1) # Extract text and bounding boxes, filtering non-alphabetic characters from text words = [] boxes = [] # Define a regular expression pattern for non-alphabetic characters non_alphabet_pattern = re.compile(r'[^a-zA-Z]+') for result in ocr_results: bbox, text, _ = result filtered_text = re.sub(non_alphabet_pattern, '', text) if filtered_text: # Only append if there are alphabetic characters left words.append(filtered_text) boxes.append([ bbox[0][0], bbox[0][1], bbox[2][0], bbox[2][1] ]) words = words[1:] def unnormalize_box(bbox, width, height): return [ width * (bbox[0] / 1000), height * (bbox[1] / 1000), width * (bbox[2] / 1000), height * (bbox[3] / 1000), ] token_boxes = encoding["bbox"].squeeze().tolist() width, height = image.size true_predictions = [model.config.id2label[pred] for pred, label in zip(predictions, labels) if label != - 100] true_labels = [model.config.id2label[label] for prediction, label in zip(predictions, labels) if label != -100] true_boxes = [unnormalize_box(box, width, height) for box, label in zip(token_boxes, labels) if label != -100] true_tokens = words # Associate languages with their levels languages_with_levels = {} current_language = None j=0 for i in range(0, len(true_labels)): if true_labels[i] == 'language': current_language = words[j] j= j+1 languages_with_levels[current_language] = true_labels[i+1] print(languages_with_levels) input_ids = encoding["input_ids"] bbox = encoding["bbox"] attention_mask = encoding["attention_mask"] st.write("Predicted labels:") # Print languages with their levels for language, level in languages_with_levels.items(): st.write(f"{language}: {level}")