app.py

import streamlit as st
import os
import numpy as np
import cv2
from PIL import Image
import torch
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from ultralytics import YOLO
import Levenshtein

# Page config
st.set_page_config(
    page_title="Thai License Plate Detection",
    page_icon="🚗",
    layout="centered"
)

# Initialize session state for models
if 'models_loaded' not in st.session_state:
    st.session_state['models_loaded'] = False

def load_ocr_models():
    """Load OCR models with proper error handling"""
    try:
        # Set environment variables to suppress warnings
        os.environ['TOKENIZERS_PARALLELISM'] = 'false'
        
        # Load processor with specific config
        processor = TrOCRProcessor.from_pretrained(
            'openthaigpt/thai-trocr',
            revision='main',
            use_auth_token=False,
            trust_remote_code=True,
            local_files_only=False
        )
        
        # Load OCR model with specific config
        ocr_model = VisionEncoderDecoderModel.from_pretrained(
            'openthaigpt/thai-trocr',
            revision='main',
            use_auth_token=False,
            trust_remote_code=True,
            local_files_only=False
        )
        
        # Move model to CPU explicitly
        ocr_model = ocr_model.to('cpu')
        
        return processor, ocr_model
    except Exception as e:
        st.error(f"Error loading OCR models: {str(e)}")
        st.error("Detailed error information:")
        import traceback
        st.code(traceback.format_exc())
        return None, None

# Load models
@st.cache_resource
def load_models():
    try:
        # Check if YOLO weights exist
        if not os.path.exists('best.pt'):
            st.error("YOLO model weights (best.pt) not found in the current directory!")
            return None, None, None

        # Load YOLO model
        try:
            yolo_model = YOLO('best.pt', task='detect')
        except Exception as yolo_error:
            st.error(f"Error loading YOLO model: {str(yolo_error)}")
            return None, None, None
        
        # Load OCR models
        processor, ocr_model = load_ocr_models()
        if processor is None or ocr_model is None:
            return None, None, None
            
        return processor, ocr_model, yolo_model
        
    except Exception as e:
        st.error(f"Error in model loading: {str(e)}")
        st.error("Detailed error information:")
        import traceback
        st.code(traceback.format_exc())
        return None, None, None

# Thai provinces list
thai_provinces = [
    "กรุงเทพมหานคร", "กระบี่", "กาญจนบุรี", "กาฬสินธุ์", "กำแพงเพชร", "ขอนแก่น", "จันทบุรี", "ฉะเชิงเทรา",
    "ชลบุรี", "ชัยนาท", "ชัยภูมิ", "ชุมพร", "เชียงราย", "เชียงใหม่", "ตรัง", "ตราด", "ตาก", "นครนายก",
    "นครปฐม", "นครพนม", "นครราชสีมา", "นครศรีธรรมราช", "นครสวรรค์", "นราธิวาส", "น่าน", "บึงกาฬ",
    "บุรีรัมย์", "ปทุมธานี", "ประจวบคีรีขันธ์", "ปราจีนบุรี", "ปัตตานี", "พะเยา", "พังงา", "พัทลุง",
    "พิจิตร", "พิษณุโลก", "เพชรบูรณ์", "เพชรบุรี", "แพร่", "ภูเก็ต", "มหาสารคาม", "มุกดาหาร", "แม่ฮ่องสอน",
    "ยโสธร", "ยะลา", "ร้อยเอ็ด", "ระนอง", "ระยอง", "ราชบุรี", "ลพบุรี", "ลำปาง", "ลำพูน", "เลย",
    "ศรีสะเกษ", "สกลนคร", "สงขลา", "สมุทรปราการ", "สมุทรสงคราม", "สมุทรสาคร", "สระแก้ว", "สระบุรี",
    "สิงห์บุรี", "สุโขทัย", "สุพรรณบุรี", "สุราษฎร์ธานี", "สุรินทร์", "หนองคาย", "หนองบัวลำภู", "อำนาจเจริญ",
    "อุดรธานี", "อุทัยธานี", "อุบลราชธานี", "อ่างทอง"
]

def get_closest_province(input_text, provinces):
    min_distance = float('inf')
    closest_province = None
    for province in provinces:
        distance = Levenshtein.distance(input_text, province)
        if distance < min_distance:
            min_distance = distance
            closest_province = province
    return closest_province, min_distance

def process_image(image, processor, ocr_model, yolo_model):
    CONF_THRESHOLD = 0.2
    data = {"plate_number": "", "province": "", "raw_province": "", "plate_crop": None, "province_crop": None}
    
    # Convert PIL Image to cv2 format
    image = np.array(image)
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    
    # Image enhancement
    lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)
    l, a, b = cv2.split(lab)
    clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8,8))
    cl = clahe.apply(l)
    enhanced = cv2.merge((cl,a,b))
    image = cv2.cvtColor(enhanced, cv2.COLOR_LAB2BGR)

    # YOLO detection
    results = yolo_model(image)
    
    # Process detections
    detections = []
    for result in results:
        for box in result.boxes:
            confidence = float(box.conf)
            class_id = int(box.cls.item())
            if confidence < CONF_THRESHOLD:
                continue
            x1, y1, x2, y2 = map(int, box.xyxy.flatten())
            detections.append((class_id, confidence, (x1, y1, x2, y2)))

    # Sort by class_id
    detections.sort(key=lambda x: x[0])
    
    for class_id, confidence, (x1, y1, x2, y2) in detections:
        cropped_image = image[y1:y2, x1:x2]
        if cropped_image.size == 0:
            continue
            
        # Preprocess for OCR
        cropped_image_gray = cv2.cvtColor(cropped_image, cv2.COLOR_BGR2GRAY)
        thresh_image = cv2.adaptiveThreshold(
            cropped_image_gray,
            255,
            cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
            cv2.THRESH_BINARY_INV,
            11,
            2
        )
        
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2,2))
        thresh_image = cv2.morphologyEx(thresh_image, cv2.MORPH_CLOSE, kernel)
        cropped_image_3d = cv2.cvtColor(thresh_image, cv2.COLOR_GRAY2RGB)
        resized_image = cv2.resize(cropped_image_3d, (128, 32))
        
        # OCR processing
        pixel_values = processor(resized_image, return_tensors="pt").pixel_values
        generated_ids = ocr_model.generate(pixel_values)
        generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
        
        # Convert crop to PIL for display
        cropped_pil = Image.fromarray(cv2.cvtColor(cropped_image, cv2.COLOR_BGR2RGB))

        if class_id == 0:  # License plate
            data["plate_number"] = generated_text
            data["plate_crop"] = cropped_pil
        elif class_id == 1:  # Province
            generated_province, distance = get_closest_province(generated_text, thai_provinces)
            data["raw_province"] = generated_text
            data["province"] = generated_province
            data["province_crop"] = cropped_pil
            
    return data

# Main app
st.title("Thai License Plate Detection 🚗")

# Load models
try:
    if not st.session_state['models_loaded']:
        with st.spinner("Loading models... (this may take a minute)"):
            processor, ocr_model, yolo_model = load_models()
            st.session_state['models_loaded'] = True
            st.session_state['processor'] = processor
            st.session_state['ocr_model'] = ocr_model
            st.session_state['yolo_model'] = yolo_model
except Exception as e:
    st.error(f"Error loading models: {str(e)}")
    st.stop()

# File uploader
uploaded_file = st.file_uploader("Upload an image of a Thai license plate", type=["jpg", "jpeg", "png"])

if uploaded_file is not None:
    try:
        # Display the uploaded image
        col1, col2 = st.columns(2)
        with col1:
            st.subheader("Uploaded Image")
            image = Image.open(uploaded_file)
            st.image(image, use_column_width=True)
        
        # Process the image
        with col2:
            st.subheader("Detection Results")
            with st.spinner("Processing image..."):
                results = process_image(
                    image,
                    st.session_state['processor'],
                    st.session_state['ocr_model'],
                    st.session_state['yolo_model']
                )
                
                if results["plate_number"]:
                    st.success("Detection successful!")
                    st.write("📝 License Plate:", results['plate_number'])
                    
                    if results['plate_crop'] is not None:
                        st.subheader("Cropped License Plate")
                        st.image(results['plate_crop'], caption="Detected License Plate Region")
                    
                    if results['raw_province']:
                        st.write("🔍 Detected Province Text:", results['raw_province'])
                        if results['province']:
                            st.write("🏠 Matched Province:", results['province'])
                        else:
                            st.write("⚠️ No close province match found")
                        
                        if results['province_crop'] is not None:
                            st.subheader("Cropped Province")
                            st.image(results['province_crop'], caption="Detected Province Region")
                    else:
                        st.write("⚠️ No province text detected")
                else:
                    st.error("No license plate detected in the image.")
                    
    except Exception as e:
        st.error(f"An error occurred: {str(e)}")

st.markdown("---")
st.markdown("### Instructions")
st.markdown("""
1. Upload an image containing a Thai license plate
2. Wait for the processing to complete
3. View the detected license plate number and province
""")

# Add footer with GitHub link
st.markdown("---")
st.markdown("Made with ❤️ by [AI Research Group KMUTT]")
st.markdown("Check out the [GitHub Repository](https://github.com/yourusername/your-repo) for more information")