app.py

#!/usr/bin/env python3
"""
YOLO Object Detection with Gradio Interface
Optimized for Hugging Face Spaces deployment
"""

import gradio as gr
import cv2
import numpy as np
from ultralytics import YOLO
from PIL import Image
import torch
import spaces
import os
import tempfile

# Global variable for models
models = {}
current_model_size = 'nano'

def load_model(model_size='nano'):
    """
    Load YOLO model based on selected size
    """
    global models, current_model_size
    
    model_names = {
        'nano': 'yolov8n.pt',
        'small': 'yolov8s.pt',
        'medium': 'yolov8m.pt',
        'large': 'yolov8l.pt',
        'xlarge': 'yolov8x.pt'
    }
    
    model_name = model_names.get(model_size, 'yolov8n.pt')
    
    # Check if model already loaded
    if model_size not in models:
        print(f"Loading {model_name}...")
        models[model_size] = YOLO(model_name)
        current_model_size = model_size
        
        # Check if CUDA is available
        if torch.cuda.is_available():
            return f"✅ Model {model_name} loaded successfully! (GPU enabled)"
        else:
            return f"✅ Model {model_name} loaded successfully! (CPU mode)"
    else:
        current_model_size = model_size
        return f"✅ Model {model_name} already loaded!"

# Use @spaces.GPU decorator for GPU functions on Hugging Face Spaces
@spaces.GPU(duration=60)
def detect_image(input_image, model_size, conf_threshold=0.25, iou_threshold=0.45):
    """
    Perform object detection on a single image
    """
    if model_size not in models:
        load_model(model_size)
    
    model = models[model_size]
    
    if input_image is None:
        return None, "No image provided"
    
    # Convert PIL Image to numpy array if necessary
    if isinstance(input_image, Image.Image):
        input_image = np.array(input_image)
    
    # Run inference
    results = model(input_image, conf=conf_threshold, iou=iou_threshold)
    
    # Get annotated image
    annotated_image = results[0].plot()
    
    # Get detection details
    detections = []
    for r in results:
        if r.boxes is not None:
            for box in r.boxes:
                if box.cls is not None:
                    class_id = int(box.cls)
                    class_name = model.names[class_id]
                    confidence = float(box.conf)
                    bbox = box.xyxy[0].tolist()
                    detections.append({
                        'class': class_name,
                        'confidence': f"{confidence:.2%}",
                        'bbox': [int(x) for x in bbox]
                    })
    
    # Create detection summary
    summary = f"Found {len(detections)} object(s)\n\n"
    if detections:
        # Count occurrences of each class
        class_counts = {}
        for det in detections:
            class_name = det['class']
            if class_name not in class_counts:
                class_counts[class_name] = 0
            class_counts[class_name] += 1
        
        summary += "Summary by class:\n"
        for class_name, count in class_counts.items():
            summary += f"  • {class_name}: {count}\n"
        
        summary += "\nDetailed detections:\n"
        for i, det in enumerate(detections, 1):
            summary += f"{i}. {det['class']} ({det['confidence']})\n"
    
    return annotated_image, summary

@spaces.GPU(duration=120)
def detect_video(input_video, model_size, conf_threshold=0.25, iou_threshold=0.45, max_frames=300):
    """
    Perform object detection on video
    """
    if model_size not in models:
        load_model(model_size)
    
    model = models[model_size]
    
    if input_video is None:
        return None, "No video provided"
    
    # Open video
    cap = cv2.VideoCapture(input_video)
    fps = int(cap.get(cv2.CAP_PROP_FPS))
    if fps == 0:
        fps = 25  # Default fallback FPS
    width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    
    # Limit processing for Spaces
    if max_frames and total_frames > max_frames:
        total_frames = max_frames
    
    # Create temporary output file
    with tempfile.NamedTemporaryFile(suffix='.mp4', delete=False) as tmp_file:
        output_path = tmp_file.name
    
    # Setup video writer
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))
    
    frame_count = 0
    detected_objects = set()
    
    # Process progress callback
    def progress_callback(current, total):
        return (current / total) if total > 0 else 0
    
    # Process video
    progress = gr.Progress()
    while cap.isOpened() and frame_count < total_frames:
        ret, frame = cap.read()
        if not ret:
            break
        
        # Run detection
        results = model(frame, conf=conf_threshold, iou=iou_threshold)
        
        # Collect detected classes
        for r in results:
            if r.boxes is not None:
                for box in r.boxes:
                    if box.cls is not None:
                        class_id = int(box.cls)
                        detected_objects.add(model.names[class_id])
        
        # Get annotated frame
        annotated_frame = results[0].plot()
        
        # Write frame
        out.write(annotated_frame)
        frame_count += 1
        
        # Update progress
        if frame_count % 10 == 0:
            progress(frame_count / total_frames, desc=f"Processing frame {frame_count}/{total_frames}")
    
    # Clean up
    cap.release()
    out.release()
    
    # Create summary
    summary = f"Processed {frame_count} frames\n"
    summary += f"Detected objects: {', '.join(sorted(detected_objects))}" if detected_objects else "No objects detected"
    
    return output_path, summary

# Create Gradio interface
def create_interface():
    with gr.Blocks(
        title="YOLO Object Detection",
        theme=gr.themes.Soft(),
        css="""
        .gradio-container {
            max-width: 1200px !important;
        }
        #title {
            text-align: center;
            margin-bottom: 1rem;
        }
        """
    ) as demo:
        gr.Markdown(
            """
            <div id="title">
            
            # 🎯 YOLO Real-Time Object Detection
            
            <p>Powered by <b>Ultralytics YOLOv8</b> - State-of-the-art object detection in your browser!</p>
            
            [![Duplicate Space](https://img.shields.io/badge/🤗%20Hugging%20Face-Duplicate%20Space-blue)](https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME?duplicate=true)
            [![Model](https://img.shields.io/badge/Model-YOLOv8-green)](https://github.com/ultralytics/ultralytics)
            [![License](https://img.shields.io/badge/License-AGPL--3.0-red)](https://github.com/ultralytics/ultralytics/blob/main/LICENSE)
            
            </div>
            """
        )
        
        # Main tabs
        with gr.Tabs() as tabs:
            # Image detection tab
            with gr.TabItem("📷 Image Detection", id=0):
                with gr.Row():
                    with gr.Column():
                        image_input = gr.Image(
                            label="Upload Image",
                            type="numpy",
                            elem_id="image_input"
                        )
                        
                        with gr.Row():
                            image_model_size = gr.Dropdown(
                                choices=['nano', 'small', 'medium', 'large', 'xlarge'],
                                value='nano',
                                label="Model Size",
                                info="Larger = more accurate but slower"
                            )
                        
                        with gr.Row():
                            image_conf = gr.Slider(
                                minimum=0.0,
                                maximum=1.0,
                                value=0.25,
                                step=0.05,
                                label="Confidence Threshold",
                                info="Higher = fewer but more confident detections"
                            )
                            image_iou = gr.Slider(
                                minimum=0.0,
                                maximum=1.0,
                                value=0.45,
                                step=0.05,
                                label="IoU Threshold",
                                info="Higher = less overlap between boxes"
                            )
                        
                        image_button = gr.Button("🔍 Detect Objects", variant="primary", size="lg")
                    
                    with gr.Column():
                        image_output = gr.Image(label="Detection Result", elem_id="image_output")
                        image_text_output = gr.Textbox(
                            label="Detection Details",
                            lines=10,
                            max_lines=20
                        )
                
                # Example images
                with gr.Row():
                    gr.Examples(
                        examples=[
                            ["https://ultralytics.com/images/bus.jpg"],
                            ["https://ultralytics.com/images/zidane.jpg"],
                        ],
                        inputs=image_input,
                        label="Try these examples"
                    )
            
            # Video detection tab
            with gr.TabItem("🎥 Video Detection", id=1):
                with gr.Row():
                    with gr.Column():
                        video_input = gr.Video(
                            label="Upload Video",
                            elem_id="video_input"
                        )
                        
                        with gr.Row():
                            video_model_size = gr.Dropdown(
                                choices=['nano', 'small', 'medium'],
                                value='nano',
                                label="Model Size",
                                info="Nano recommended for videos"
                            )
                        
                        with gr.Row():
                            video_conf = gr.Slider(
                                minimum=0.0,
                                maximum=1.0,
                                value=0.25,
                                step=0.05,
                                label="Confidence Threshold"
                            )
                            video_iou = gr.Slider(
                                minimum=0.0,
                                maximum=1.0,
                                value=0.45,
                                step=0.05,
                                label="IoU Threshold"
                            )
                        
                        max_frames = gr.Slider(
                            minimum=10,
                            maximum=300,
                            value=100,
                            step=10,
                            label="Max Frames to Process",
                            info="Limit for Spaces resources"
                        )
                        
                        video_button = gr.Button("🎬 Process Video", variant="primary", size="lg")
                    
                    with gr.Column():
                        video_output = gr.Video(
                            label="Processed Video",
                            elem_id="video_output"
                        )
                        video_text_output = gr.Textbox(
                            label="Processing Summary",
                            lines=4
                        )
            
            # About tab
            with gr.TabItem("ℹ️ About", id=2):
                gr.Markdown(
                    """
                    ## About YOLO (You Only Look Once)
                    
                    YOLO is a state-of-the-art, real-time object detection system. This app uses **YOLOv8** from Ultralytics,
                    the latest evolution building on Joseph Redmon's original YOLO architecture.
                    
                    ### 🚀 Model Sizes
                    
                    | Model | Parameters | Speed (CPU) | mAP | Use Case |
                    |-------|-----------|-------------|-----|----------|
                    | Nano | 3.2M | ~100ms | 37.3 | Real-time, edge devices |
                    | Small | 11.2M | ~200ms | 44.9 | Balanced performance |
                    | Medium | 25.9M | ~400ms | 50.2 | Good accuracy |
                    | Large | 43.7M | ~800ms | 52.9 | High accuracy |
                    | XLarge | 68.2M | ~1600ms | 53.9 | Best accuracy |
                    
                    ### 🎯 Detectable Objects (COCO Dataset)
                    
                    YOLOv8 can detect 80 different object classes including:
                    - **People**: person
                    - **Vehicles**: bicycle, car, motorcycle, airplane, bus, train, truck, boat
                    - **Animals**: bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe
                    - **Sports**: frisbee, skis, snowboard, sports ball, kite, baseball bat, skateboard, surfboard, tennis racket
                    - **Food**: banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake
                    - **Household**: chair, couch, bed, dining table, toilet, TV, laptop, mouse, keyboard, cell phone, book, clock
                    - And many more!
                    
                    ### 📖 Resources
                    
                    - [Ultralytics YOLOv8 Documentation](https://docs.ultralytics.com/)
                    - [Original YOLO Paper](https://arxiv.org/abs/1506.02640)
                    - [GitHub Repository](https://github.com/ultralytics/ultralytics)
                    
                    ### 🤝 Credits
                    
                    - Original YOLO by Joseph Redmon
                    - YOLOv8 by Ultralytics
                    - Gradio by Hugging Face
                    - Deployed on Hugging Face Spaces
                    
                    ---
                    
                    Made with ❤️ using Gradio and Ultralytics
                    """
                )
        
        # Event handlers
        image_button.click(
            fn=detect_image,
            inputs=[image_input, image_model_size, image_conf, image_iou],
            outputs=[image_output, image_text_output]
        )
        
        video_button.click(
            fn=detect_video,
            inputs=[video_input, video_model_size, video_conf, video_iou, max_frames],
            outputs=[video_output, video_text_output]
        )
        
        # Load initial model on startup
        demo.load(
            fn=lambda: load_model('nano'),
            inputs=None,
            outputs=None
        )
    
    return demo

# Main execution
if __name__ == "__main__":
    # Create and launch interface
    demo = create_interface()
    demo.queue()  # Enable queue for better performance
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        show_error=True
    )