pdfextract_fun.py

import os
import re
import warnings
import cv2
import fitz  # PyMuPDF
import numpy as np
import pytesseract
import torch
from PIL import Image
from tqdm import tqdm
from detectron2.config import get_cfg
from detectron2.data import MetadataCatalog
from detectron2.engine import DefaultPredictor
from detectron2.utils.visualizer import ColorMode, Visualizer
from unilm.dit.object_detection.ditod import add_vit_config

# Filter specific warnings
warnings.filterwarnings("ignore", message="None of the inputs have requires_grad=True. Gradients will be None")
warnings.filterwarnings("ignore", message="torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument.")

# Configuration setup
def setup_config():
    cfg = get_cfg()
    add_vit_config(cfg)
    cfg.merge_from_file("cascade_dit_base.yml")
    cfg.MODEL.WEIGHTS = "publaynet_dit-b_cascade.pth"
    cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
    return cfg

# Analyze image
def analyze_image(img, cfg):
    """Analyze an image and return the result image, output, and visualizer."""
    md = MetadataCatalog.get(cfg.DATASETS.TEST[0])
    thing_classes = ["table"] if cfg.DATASETS.TEST[0] == 'icdar2019_test' else ["text", "title", "list", "table", "figure"]
    md.set(thing_classes=thing_classes)

    output = DefaultPredictor(cfg)(img)["instances"]
    v = Visualizer(img[:, :, ::-1], metadata=md, scale=1.0, instance_mode=ColorMode.SEGMENTATION)
    result = v.draw_instance_predictions(output.to("cpu"))
    return result.get_image()[:, :, ::-1], output, v

# PDF to JPEG conversion
def convert_pdf_to_jpg(pdf_path, output_folder, zoom_factor=2):
    """Convert PDF file to JPEG images, saved in the specified output folder."""
    doc = fitz.open(pdf_path)
    for page_num, page in enumerate(doc):
        mat = fitz.Matrix(zoom_factor, zoom_factor)
        pix = page.get_pixmap(matrix=mat)
        output_file = os.path.join(output_folder, f"page_{page_num}.jpg")
        pix.save(output_file)

# Process JPEG images in a folder
def process_jpeg_images((output_folder, cfg, batch_size=10):
    image_paths = [os.path.join(output_folder, f) for f in os.listdir(output_folder) if f.endswith('.jpg')]
    batches = [image_paths[i:i + batch_size] for i in range(0, len(image_paths), batch_size)]
    
    for batch in tqdm(batches, desc="Processing images in batches"):
        images = [cv2.imread(image_path) for image_path in batch]
        batch_results = batch_analyze_images(images, cfg)  # This function needs to be implemented to support batch processing
        
        for i, (result_image, output, v) in enumerate(batch_results):
            # Assuming batch_analyze_images returns a list of tuples, each containing the results for one image
            save_extracted_instances(images[i], output, i, output_folder)

# Save extracted instances
def save_extracted_instances(img, output, page_num, dest_folder, confidence_threshold=0.8):
    """Save instances extracted from an image to the destination folder."""
    class_names = {0: "text", 1: "title", 2: "list", 3: "table", 4: "figure"}
    instances = output.to("cpu")
    for i, (box, class_id, score) in enumerate(zip(instances.pred_boxes.tensor.numpy(), instances.pred_classes.tolist(), instances.scores.tolist())):
        if score >= confidence_threshold and class_names.get(class_id) in ["figure", "table", "text"]:
            x1, y1, x2, y2 = map(int, box)
            cropped_image = img[y1:y2, x1:x2]
            if np.std(cropped_image) > 0 and (y2 - y1) > 0:  # Replace with actual thresholds if needed
                save_path = os.path.join(dest_folder, f"page_{page_num}_{class_names[class_id]}_{i + 1}.jpg")
                cv2.imwrite(save_path, cropped_image)

# Additional functions like delete_files_in_folder, rename_files_sequentially, ocr_folder, and ocr_image can be included as is, assuming they were satisfactory before.

cfg = setup_config()
# Example usage
convert_pdf_to_jpg("sample.pdf", "output_folder")
process_jpeg_images("output_folder", cfg)