# Smart Parking System - Hugging Face Spaces Version
# Run this in Hugging Face Spaces

!pip install -q torch torchvision transformers pillow matplotlib numpy opencv-python-headless timm einops gradio
!pip install -q git+https://github.com/facebookresearch/segment-anything-2.git
!pip install -q ultralytics supervision

import torch
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.patches as patches
from datetime import datetime, timedelta
import cv2
from transformers import AutoProcessor, AutoModelForCausalLM
import gradio as gr
import warnings
warnings.filterwarnings('ignore')


class StateOfTheArtParkingDetector:
    def __init__(self, total_spaces=50):
        self.total_spaces = total_spaces
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        
        print(f"🚀 Loading Models... Device: {self.device}")

        # Try Florence-2 (optional, can skip if CPU)
        self.florence_available = False
        try:
            self.processor = AutoProcessor.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True)
            self.model = AutoModelForCausalLM.from_pretrained(
                "microsoft/Florence-2-base",
                torch_dtype=torch.float16 if self.device=="cuda" else torch.float32,
                trust_remote_code=True
            ).to(self.device)
            self.florence_available = True
            print("✅ Florence-2 loaded")
        except:
            print("⚠️ Florence-2 unavailable, skipping")

        # YOLO-World
        try:
            from ultralytics import YOLO
            self.yolo_world = YOLO('yolov8x-worldv2.pt')
            self.yolo_world_available = True
            print("✅ YOLO-World loaded")
        except Exception as e:
            print(f"⚠️ YOLO-World unavailable: {e}")
            self.yolo_world_available = False

    # Florence detection
    def detect_with_florence(self, image):
        if not self.florence_available:
            return []
        # Minimal placeholder to avoid crash
        return []

    # YOLO-World detection
    def detect_with_yolo_world(self, image):
        if not self.yolo_world_available:
            return []
        self.yolo_world.set_classes([
            "car", "vehicle", "automobile", "truck", "van",
            "SUV", "sedan", "parked car", "parking space with car"
        ])
        img_array = np.array(image)
        results = self.yolo_world(img_array, conf=0.15, iou=0.3, verbose=False)
        detections = []
        for result in results:
            boxes = result.boxes
            for box in boxes:
                x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
                conf = float(box.conf[0])
                detections.append({'box':[int(x1), int(y1), int(x2), int(y2)],
                                   'label':'car', 'score':conf, 'method':'yolo-world'})
        return detections

    # Simple CV detection fallback
    def advanced_cv_detection(self, image):
        img_array = np.array(image)
        gray = cv2.cvtColor(img_array, cv2.COLOR_RGB2GRAY)
        edges = cv2.Canny(gray, 30, 100)
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (7,7))
        closed = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel, iterations=2)
        contours, _ = cv2.findContours(closed, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        h, w = gray.shape
        min_area, max_area = (h*w)//600, (h*w)//20
        detections = []
        for contour in contours:
            area = cv2.contourArea(contour)
            if min_area < area < max_area:
                x, y, cw, ch = cv2.boundingRect(contour)
                aspect_ratio = cw/ch if ch>0 else 0
                if 0.4<aspect_ratio<3.5:
                    hull = cv2.convexHull(contour)
                    hull_area = cv2.contourArea(hull)
                    solidity = area/hull_area if hull_area>0 else 0
                    if solidity > 0.4:
                        detections.append({'box':[x, y, x+cw, y+ch], 'label':'car', 'score':0.7, 'method':'cv'})
        return detections

    # Merge detections (NMS)
    def merge_all_detections(self, all_detections):
        if len(all_detections)==0: return []
        normalized = []
        for det in all_detections:
            x1, y1, x2, y2 = det['box']
            normalized.append({'box':[x1,y1,x2,y2], 'score':det['score'], 'method':det.get('method','unknown')})
        normalized.sort(key=lambda x: x['score'], reverse=True)
        keep=[]
        while normalized:
            best = normalized.pop(0)
            keep.append(best)
            normalized = [d for d in normalized if self.calculate_iou(best['box'], d['box'])<0.5]
        return keep

    def calculate_iou(self, box1, box2):
        x1_1, y1_1, x2_1, y2_1 = box1
        x1_2, y1_2, x2_2, y2_2 = box2
        x1_i, y1_i = max(x1_1,x1_2), max(y1_1,y1_2)
        x2_i, y2_i = min(x2_1,x2_2), min(y2_1,y2_2)
        if x2_i<x1_i or y2_i<y1_i: return 0.0
        inter = (x2_i-x1_i)*(y2_i-y1_i)
        area1 = (x2_1-x1_1)*(y2_1-y1_1)
        area2 = (x2_2-x1_2)*(y2_2-y1_2)
        union = area1+area2-inter
        return inter/union if union>0 else 0

    def create_annotated_image(self, image, detections):
        img_array = np.array(image)
        fig, ax = plt.subplots(1, figsize=(12,8))
        ax.imshow(img_array)
        for i, det in enumerate(detections, 1):
            x1, y1, x2, y2 = det['box']
            rect = patches.Rectangle((x1,y1,x2-x1,y2-y1), linewidth=3, edgecolor='#00ff00', facecolor='none')
            ax.add_patch(rect)
            ax.text(x1+(x2-x1)/2, y1+(y2-y1)/2, f"{i}", 
                    bbox=dict(facecolor='#00ff00', alpha=0.9, boxstyle='circle,pad=0.3'),
                    fontsize=12, color='black', weight='bold', ha='center', va='center')
        ax.axis('off')
        plt.tight_layout()
        fig.canvas.draw()
        img_plot = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
        img_plot = img_plot.reshape(fig.canvas.get_width_height()[::-1]+(3,))
        plt.close(fig)
        return Image.fromarray(img_plot)

    # --- Main function for Gradio ---
    def process_image(self, image, total_spaces):
        if image is None: return None, "Upload an image", ""
        self.total_spaces = total_spaces
        if isinstance(image, np.ndarray): image = Image.fromarray(image)
        all_dets=[]
        all_dets.extend(self.detect_with_florence(image))
        all_dets.extend(self.detect_with_yolo_world(image))
        all_dets.extend(self.advanced_cv_detection(image))
        final_dets = self.merge_all_detections(all_dets)
        count = len(final_dets)
        annotated_img = self.create_annotated_image(image, final_dets)
        occupancy_rate = min((count/total_spaces)*100, 100)
        available = max(total_spaces-count,0)
        status = "🟢 AVAILABLE" if occupancy_rate<50 else "🟡 BUSY" if occupancy_rate<70 else "🟠 CRITICAL" if occupancy_rate<90 else "🔴 FULL"
        pred_time = (datetime.now()+timedelta(hours=available/5)).strftime('%I:%M %p')
        status_html = f"<h2 style='color:green'>{status}</h2>"
        stats_text = f"Total Vehicles: {count}\nAvailable: {available}\nOccupancy: {occupancy_rate:.1f}%"
        return annotated_img, status_html, stats_text


# --- Initialize Detector ---
detector = StateOfTheArtParkingDetector()

# --- Gradio Interface ---
with gr.Blocks() as demo:
    gr.Markdown("## 🅿️ Smart Parking Management System")
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(label="Parking Lot Image", type="pil")
            total_spaces_slider = gr.Slider(10, 200, value=50, step=5, label="Total Spaces")
            analyze_btn = gr.Button("Analyze Parking Lot")
        with gr.Column():
            output_image = gr.Image(label="Detected Vehicles", type="pil")
            status_html = gr.HTML(label="Parking Status")
    stats_output = gr.Textbox(label="Detailed Stats", lines=10)
    analyze_btn.click(fn=detector.process_image, inputs=[input_image,total_spaces_slider], outputs=[output_image,status_html,stats_output])

demo.launch()