scripts/model.py

import os
import json
import numpy as np
from PIL import Image
import pandas as pd
from IPython.display import Image
from ultralytics import YOLO
import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer, Trainer, TrainingArguments
from datasets import load_dataset
import cv2
import pytesseract
from PIL import Image, ImageEnhance
import numpy as np
from sklearn.metrics import precision_score, recall_score, f1_score


def ocr_core(image):
    """
    Run Tesseract OCR on the preprocessed image and return the extracted text.

    Inputs:
        image (PIL.Image): The preprocessed image to run OCR on.

    Returns:
        str: The text extracted from the image.
    
    """
    data = pytesseract.image_to_data(image, output_type=pytesseract.Output.DICT)
    df = pd.DataFrame(data)
    df = df[df['conf'] != -1]
    df['left_diff'] = df.groupby('block_num')['left'].diff().fillna(0).astype(int)
    df['prev_width'] = df['width'].shift(1).fillna(0).astype(int)
    df['spacing'] = (df['left_diff'] - df['prev_width']).fillna(0).astype(int)
    df['text'] = df.apply(lambda x: '\n' + x['text'] if (x['word_num'] == 1) & (x['block_num'] != 1) else x['text'], axis=1)
    df['text'] = df.apply(lambda x: ',' + x['text'] if x['spacing'] > 100 else x['text'], axis=1)
    ocr_text = ""
    for text in df['text']:
        ocr_text += text + ' '
    return ocr_text

def improve_ocr_accuracy(img):
    """
    Preprocess the image to improve OCR accuracy by resizing, increasing contrast, and thresholding.

    Inputs:
        img (str): The path to the image file.

    Returns:
        np.ndarray: The preprocessed image as a binary thresholded array.
    
    """
    # Read image with PIL (for color preservation)
    img = Image.open(img)
    
    # Increase image size (can improve accuracy for small text)
    img = img.resize((img.width * 4, img.height * 4))
    
    # Increase contrast
    enhancer = ImageEnhance.Contrast(img)
    img = enhancer.enhance(2)

    _, thresh = cv2.threshold(np.array(img), 127, 255, cv2.THRESH_BINARY_INV)
    
    return thresh

def create_ocr_outputs():
    """
    
    Process images in a directory, run OCR on them, and save the extracted text 
    to corresponding text files.
    
    Inputs:

    Returns:

    """
    directory_path = os.getcwd() + '/data/processed/hand_labeled_tables/hand_labeled_tables'

    for root, dirs, files in os.walk(directory_path):
        # Print the current directory
        print(f"Current directory: {root}")
        
        # Print all subdirectories in the current directory
        print("Subdirectories:")
        for dir in dirs:
            print(f"- {dir}")
        
        # Print all files in the current directory
        print("Files:")
        for image_path in files:
            print(f"- {image_path}")
            full_path = os.path.join(root, image_path)
            # Preprocess the image
            preprocessed_image = improve_ocr_accuracy(full_path)

            ocr_text = ocr_core(preprocessed_image)
            with open(os.getcwd() + f"/data/processed/annotations/{image_path.split('.')[0]}.txt", 'wb') as f:
                f.write(ocr_text.encode('utf-8'))
        
        print("\n")  # Add a blank line for readability

def prepare_dataset(ocr_dir, csv_dir, output_file):
    """
    Prepare a dataset by combining OCR text files and corresponding CSV 
    files into a JSONL format.

    Inputs:
        ocr_dir (str): The directory containing OCR text files.
        csv_dir (str): The directory containing CSV files.
        output_file (str): The path to the output JSONL file.
    
    Returns:
    
    """
    with open(output_file, 'w', encoding='utf-8') as jsonl_file:
        for filename in os.listdir(ocr_dir):
            if filename.endswith('.txt'):
                ocr_path = os.path.join(ocr_dir, filename)
                csv_path = os.path.join(csv_dir, filename)
                print(csv_path)
                
                with open(ocr_path, 'r', encoding='utf-8') as ocr_file:
                    ocr_text = ocr_file.read()
                
                with open(csv_path, 'r', encoding='utf-8') as csv_file:
                    csv_text = csv_file.read()
                
                json_object = {
                    "prompt": ocr_text,
                    "completion": csv_text
                }
                jsonl_file.write(json.dumps(json_object) + '\n')

def tokenize_function(examples):
    """
    Tokenize the inputs and create labels for the tokenized inputs.

    Inputs:
        examples (dict): A dictionary containing 'prompt' and 'completion' keys.

    Returns:
        dict: A dictionary containing tokenized inputs and labels.
    
    """
    # Tokenize the inputs
    inputs = tokenizer(examples['prompt'], truncation=True, padding='max_length', max_length=1012)
    
    # Create labels which are the same as input_ids
    inputs['labels'] = inputs['input_ids'].copy()
    return inputs

def calculate_metrics(model, tokenizer, texts, labels):
    """
    Calculate evaluation metrics for the model based on the provided texts and labels.

    Inputs:
        model (GPT2LMHeadModel): The language model to evaluate.
        tokenizer (GPT2Tokenizer): The tokenizer for the model.
        texts (list): A list of input texts.
        labels (list): A list of corresponding labels.

    Returns:
        list: A list containing precision, recall, F1 score.
    
    """
    model.eval()
    all_predictions = []
    all_labels = []
    total_loss = 0
    total_tokens = 0
    
    with torch.no_grad():
        for text, label in zip(texts, labels):
            # Tokenize input and label
            input_ids = tokenizer.encode(text, return_tensors="pt")
            label_ids = tokenizer.encode(label, return_tensors="pt")[0]
            
            # Generate prediction
            output = model.generate(input_ids, max_length=input_ids.shape[1] + len(label_ids), num_return_sequences=1)
            predicted_ids = output[0][input_ids.shape[1]:]
            
            # Convert ids to tokens
            predicted_tokens = tokenizer.convert_ids_to_tokens(predicted_ids)
            label_tokens = tokenizer.convert_ids_to_tokens(label_ids)
            
            # Extend predictions and labels
            all_predictions.extend(predicted_tokens)
            all_labels.extend(label_tokens)
            
            # Calculate loss
            outputs = model(input_ids=input_ids, labels=label_ids.unsqueeze(0))
            loss = outputs.loss
            total_loss += loss.item() * len(label_ids)
            total_tokens += len(label_ids)
    
    # Calculate metrics
    precision = precision_score(all_labels, all_predictions, average='weighted', zero_division=0)
    recall = recall_score(all_labels, all_predictions, average='weighted', zero_division=0)
    f1 = f1_score(all_labels, all_predictions, average='weighted', zero_division=0)
    
    return precision, recall, f1


if __name__ == '__main__':
    
    # Ensure you have installed Tesseract OCR and set the path
    pytesseract.pytesseract.tesseract_cmd = r'C:/Program Files/Tesseract-OCR/tesseract.exe'  # Update this path for your system

    # Ensure CUDA is available
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"Using device: {device}")

    # Load a pretrained YOLOv8 model
    model = YOLO('yolov8l.pt')

    # Train the model on your custom dataset
    results = model.train(
        data='config.yaml',
        epochs=1,
        imgsz=640,
        batch=8,
        name='yolov8l_custom',
        device=device
    )

    metrics = model.val()
    print(metrics.box.map)  
    
    results = model.val()

    model.save(os.getcwd() + '/models/trained_yolov8.pt')
    
    create_ocr_outputs()

    # Usage
    ocr_dir = os.getcwd() + '/data/processed/annotations'
    csv_dir = os.getcwd() + '/data/processed/hand_labeled_tables'
    output_file = 'dataset.jsonl'
    prepare_dataset(ocr_dir, csv_dir, output_file)
    

    # Load the dataset
    dataset = load_dataset('json', data_files={'train': 'dataset.jsonl'})
    dataset = dataset['train'].train_test_split(test_size=0.1)

    tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
    tokenizer.add_special_tokens({'pad_token': '[PAD]'})

    tokenized_dataset = dataset.map(tokenize_function, batched=True)

    gpt_model = GPT2LMHeadModel.from_pretrained('gpt2')

    gpt_model.resize_token_embeddings(len(tokenizer))

    training_args = TrainingArguments(
        output_dir='./results',
        num_train_epochs=3,
        per_device_train_batch_size=2,
        per_device_eval_batch_size=2,
        warmup_steps=500,
        weight_decay=0.01,
        logging_dir='./logs',
        logging_steps=10,
        evaluation_strategy="epoch",  # Evaluate at the end of each epoch
        save_strategy="epoch",  # Save at the end of each epoch
        load_best_model_at_end=True,  # Load the best model when finished training (based on evaluation)
        metric_for_best_model="eval_loss",  # Use eval_loss to determine the best model
    )

    # Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_dataset['train'],
        eval_dataset=tokenized_dataset['test'],
    )

    # Train the model
    trainer.train()

    # Evaluate the model
    eval_results = trainer.evaluate()
    print(f"Evaluation results: {eval_results}")

    # Save the model
    gpt_model.save_pretrained(os.getcwd() + '/models/gpt')
    tokenizer.save_pretrained(os.getcwd() + '/models/gpt')
    
    # Calculate metrics
    precision, recall, f1 = calculate_metrics(gpt_model, tokenizer, dataset['test']['text'], dataset['test']['label'])

    # Display metrics
    print(f"Precision: {precision:.4f}")
    print(f"Recall: {recall:.4f}")
    print(f"F1 Score: {f1:.4f}")