app_main.py

from flask import Flask, render_template, Response, flash, redirect, url_for, request, jsonify 
import cv2
import numpy as np
from unstructured.partition.pdf import partition_pdf
import json, base64, io, os
from PIL import Image, ImageEnhance, ImageDraw
from imutils.perspective import four_point_transform
from dotenv import load_dotenv
import pytesseract
from transformers import AutoProcessor, AutoModelForImageTextToText
from langchain_community.document_loaders.image_captions import ImageCaptionLoader

app = Flask(__name__)

pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
poppler_path=r"C:\poppler-23.11.0\Library\bin"

count = 0
PDF_GET = r"E:\Pratham\2025\Harsh Sir\Scratch Vision\images\scratch_crab.pdf"

OUTPUT_FOLDER = "OUTPUTS"
DETECTED_IMAGE_FOLDER_PATH = os.path.join(OUTPUT_FOLDER,"DETECTED_IMAGE")
IMAGE_FOLDER_PATH = os.path.join(OUTPUT_FOLDER, "SCANNED_IMAGE")
JSON_FOLDER_PATH = os.path.join(OUTPUT_FOLDER, "EXTRACTED_JSON")

for path in [OUTPUT_FOLDER, IMAGE_FOLDER_PATH, DETECTED_IMAGE_FOLDER_PATH, JSON_FOLDER_PATH]:
    os.makedirs(path, exist_ok=True)

# Model Initialization
smolvlm256m_processor = AutoProcessor.from_pretrained("HuggingFaceTB/SmolVLM-256M-Instruct")
smolvlm256m_model = AutoModelForImageTextToText.from_pretrained("HuggingFaceTB/SmolVLM-256M-Instruct").to("cpu")

# SmolVLM Image Captioning functioning
def get_smolvlm_caption(image: Image.Image, prompt: str = "") -> str:
    # Ensure exactly one <image> token
    if "<image>" not in prompt:
        prompt = f"<image> {prompt.strip()}" 
    
    num_image_tokens = prompt.count("<image>")
    if num_image_tokens != 1:
        raise ValueError(f"Prompt must contain exactly 1 <image> token. Found {num_image_tokens}")
    
    inputs = smolvlm256m_processor(images=[image], text=[prompt], return_tensors="pt").to("cpu")
    output_ids = smolvlm256m_model.generate(**inputs, max_new_tokens=100)
    return smolvlm256m_processor.decode(output_ids[0], skip_special_tokens=True)

# --- FUNCTION: Extract images from saved PDF ---
def extract_images_from_pdf(pdf_path, output_json_path):
    ''' Extract images from PDF and generate structured sprite JSON '''
    
    pdf_filename = os.path.splitext(os.path.basename(pdf_path))[0]  # e.g., "scratch_crab"
    pdf_dir_path = os.path.dirname(pdf_path).replace("/", "\\")
    
    # Create subfolders
    extracted_image_subdir = os.path.join(DETECTED_IMAGE_FOLDER_PATH, pdf_filename)
    json_subdir = os.path.join(JSON_FOLDER_PATH, pdf_filename)
    os.makedirs(extracted_image_subdir, exist_ok=True)
    os.makedirs(json_subdir, exist_ok=True)
    
    # Output paths
    output_json_path = os.path.join(json_subdir, "extracted.json")
    final_json_path = os.path.join(json_subdir, "extracted_sprites.json")
    
    elements = partition_pdf(
        filename=pdf_path,
        strategy="hi_res",
        extract_image_block_types=["Image"], 
        extract_image_block_to_payload=True,  # Set to True to get base64 in output
    )
    
    with open(output_json_path, "w") as f:
        json.dump([element.to_dict() for element in elements], f, indent=4)

    # Display extracted images
    with open(output_json_path, 'r') as file:
        file_elements = json.load(file)
        
    # extracted_images_dir = os.path.join(os.path.dirname(output_json_path), "extracted_images")
    # os.makedirs(extracted_images_dir, exist_ok=True)
    
     # Prepare manipulated sprite JSON structure
    manipulated_json = {}

    # Final manipulated file (for captions)
    final_json_path = output_json_path.replace(".json", "_sprites.json")
    
    # If JSON already exists, load it and find the next available Sprite number
    if os.path.exists(final_json_path):
        with open(final_json_path, "r") as existing_file:
            manipulated = json.load(existing_file)
        # Determine the next available index (e.g., Sprite 4 if 1–3 already exist)
        existing_keys = [int(k.replace("Sprite ", "")) for k in manipulated.keys()]
        start_count = max(existing_keys, default=0) + 1
    else:
        start_count = 1
    
    sprite_count = start_count
    for i,element in enumerate(file_elements):
        if "image_base64" in element["metadata"]:
            image_data = base64.b64decode(element["metadata"]["image_base64"])
            image = Image.open(io.BytesIO(image_data)).convert("RGB")
            image.show(title=f"Extracted Image {i+1}")
            image_path = os.path.join(extracted_image_subdir, f"Sprite_{i+1}.png")
            image.save(image_path)
            
            description = get_smolvlm_caption(image, prompt="Give a brief Description")
            name = get_smolvlm_caption(image, prompt="give a short name/title of this Image.")
            
            manipulated_json[f"Sprite {sprite_count}"] = {
                "name": name,
                "base64": element["metadata"]["image_base64"],
                "file-path": pdf_dir_path,
                "description":description
            }
            sprite_count += 1

# Save manipulated JSON
    with open(final_json_path, "w") as sprite_file:
        json.dump(manipulated_json, sprite_file, indent=4)

    print(f"✅ Manipulated sprite JSON saved: {final_json_path}")
    return final_json_path, manipulated_json

# API endpoint
@app.route('/process_static_pdf', methods=['POST'])
def process_static_pdf():
    # Option 1: Use hardcoded static PDF
    pdf_path = r"E:\Pratham\2025\Harsh Sir\Scratch Vision\images\scratch_crab.pdf"

    # Optional: Allow override via JSON request body
    if request.json and "pdf_path" in request.json:
        pdf_path = request.json["pdf_path"]

    if not os.path.isfile(pdf_path):
        return jsonify({"error": f"File not found: {pdf_path}"}), 400

    # json_path = os.path.join(JSON_FOLDER_PATH, "extracted.json")
    json_path = None
    output_path, result = extract_images_from_pdf(pdf_path, json_path)
    
    return jsonify({
        "message": "✅ PDF processed successfully",
        "output_json": output_path,
        "sprites": result
    })

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=7860, debug=True)