Upload 4 files

README.md

@@ -1,14 +1,59 @@
----
-title: License Plate
-emoji: 😻
-colorFrom: pink
-colorTo: green
-sdk: streamlit
-sdk_version: 1.41.1
-app_file: app.py
-pinned: false
-license: mit
-short_description: license-late
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Thai License Plate Detection App 🚗
+
+This Streamlit application detects and recognizes Thai license plates and provinces from images. It uses YOLOv8 for object detection and TrOCR for text recognition.
+
+## Features
+
+- 📷 Upload images containing Thai license plates
+- 🔍 Detect and extract license plate numbers
+- 🏠 Recognize and match province names
+- 🖼️ Display cropped regions of detected plates and provinces
+- 🎯 High accuracy text recognition using TrOCR
+
+## How to Use
+
+1. Upload an image containing a Thai license plate using the file uploader
+2. Wait for the processing to complete
+3. View the results:
+   - Detected license plate number
+   - Cropped license plate image
+   - Detected province name
+   - Cropped province image
+
+## Technical Details
+
+The application uses:
+- YOLOv8 for license plate and province detection
+- TrOCR (Thai) for text recognition
+- OpenCV for image preprocessing
+- Levenshtein distance for province name matching
+
+## Models
+
+- Object Detection: YOLOv8 (custom trained for Thai license plates)
+- Text Recognition: openthaigpt/thai-trocr
+
+## Deployment
+
+This app is deployed on Hugging Face Spaces. The deployment includes:
+- Streamlit web interface
+- Pre-trained YOLO model weights
+- Required Python dependencies
+
+## Requirements
+
+All required packages are listed in `requirements.txt`. The main dependencies are:
+- streamlit
+- opencv-python-headless
+- transformers
+- ultralytics
+- torch
+- python-Levenshtein
+
+## License
+
+[Your chosen license]
+
+## Credits
+
+Created by [Your Name/Organization] 

app.py

@@ -0,0 +1,201 @@
+import streamlit as st
+import os
+import numpy as np
+import cv2
+from PIL import Image
+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
+
+# Load models
+@st.cache_resource
+def load_models():
+    processor = TrOCRProcessor.from_pretrained('openthaigpt/thai-trocr')
+    ocr_model = VisionEncoderDecoderModel.from_pretrained('openthaigpt/thai-trocr')
+    yolo_model = YOLO('best.pt')  # Make sure to include this in the repository
+    return processor, ocr_model, yolo_model
+
+# 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 [Your Name/Organization]")
+st.markdown("Check out the [GitHub Repository](https://github.com/yourusername/your-repo) for more information") 

best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b1da8d9362a1005aa5b060b0ac53b4622677e753eded2893da10b6a69bc9fb7
+size 5468691

requirements.txt

@@ -0,0 +1,8 @@
+streamlit==1.29.0
+opencv-python-headless==4.8.1.78
+numpy==1.26.2
+Pillow==10.1.0
+transformers==4.36.2
+torch==2.1.2
+ultralytics==8.0.227
+python-Levenshtein==0.23.0