app.py

import gradio as gr
from ultralytics import YOLO
from PIL import ImageDraw, ImageEnhance, ImageFilter, Image
import numpy as np
import pandas as pd
from datetime import datetime
import os

# Load the model
model = YOLO('best.pt')

def preprocess_image(image, blur_amount, brightness_level, rotation_angle):
    # Convert to PIL Image if needed
    if isinstance(image, np.ndarray):
        image = Image.fromarray(image)
    
    # Apply brightness adjustment
    enhancer = ImageEnhance.Brightness(image)
    image = enhancer.enhance(brightness_level)
    
    # Apply blur
    if blur_amount > 0:
        image = image.filter(ImageFilter.GaussianBlur(radius=blur_amount))
    
    # Apply rotation with white background
    if rotation_angle != 0:
        # Convert to RGBA to handle transparency
        if image.mode != 'RGBA':
            image = image.convert('RGBA')
        
        
        rotated = image.rotate(rotation_angle, expand=True, fillcolor=(255, 255, 255, 255))
        
        # Convert back to RGB
        image = Image.new('RGB', rotated.size, (255, 255, 255))
        image.paste(rotated, mask=rotated.split()[3]) 
    
    return image

def preview_update(image, blur_amount, brightness_level, rotation_angle):
    if image is None:
        return None
    return preprocess_image(image, blur_amount, brightness_level, rotation_angle)

def detect_objects(image, confidence, blur_amount, brightness_level, rotation_angle):
    if image is None:
        return None, None
        
    # First preprocess the image
    processed_image = preprocess_image(image, blur_amount, brightness_level, rotation_angle)
    
    # Perform object detection
    results = model(processed_image, conf=confidence)
    
    # Draw bounding boxes
    img = processed_image.copy()
    draw = ImageDraw.Draw(img)
    
    # Prepare detection data for table
    detection_data = []
    for result in results:
        for box in result.boxes:
            xmin, ymin, xmax, ymax = [int(val) for val in box.xyxy[0]]
            confidence_score = float(box.conf[0])
            # Draw rectangle with thicker border
            draw.rectangle([(xmin, ymin), (xmax, ymax)], outline='green', width=8)
            draw.text((xmin, ymin-20), f'{confidence_score:.2f}', fill='green', font_size=20)
            
            # Add to detection data
            detection_data.append([
                f"{confidence_score:.2f}",
                f"({xmin}, {ymin})",
                f"({xmax}, {ymax})"
            ])
    
    # If no detections, return empty list for table to render properly
    if not detection_data:
        detection_data = []
    
    return img, detection_data

def process_multiple_images(images, confidence, blur_amount, brightness_level, rotation_angle):
    if not images:
        return None, None
    
    all_results = []
    all_data = []
    
    for idx, file in enumerate(images):
        # Open image from file
        img = Image.open(file.name)
        result, data = detect_objects(img, confidence, blur_amount, brightness_level, rotation_angle)
        if result is not None:
            all_results.append(result)
            # Add image number to each detection in data
            image_data = [[f"Image {idx+1}"] + row for row in data]
            all_data.extend(image_data)
    
    return all_results, all_data

def preview_multiple(files, blur_amount, brightness_level, rotation_angle):
    if not files:
        return None
    
    previews = []
    for file in files:
        img = Image.open(file.name)
        preview = preview_update(img, blur_amount, brightness_level, rotation_angle)
        if preview is not None:
            previews.append(preview)
    
    return previews


def export_to_csv(data):
    if data is None or data.empty:
        return None
    
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    filename = f"detection_results_{timestamp}.csv"
    
    temp_path = os.path.join(os.getcwd(), filename)
    data.to_csv(temp_path, index=False)
    
    return temp_path


def cleanup_temp_files():
    for file in os.listdir():
        if file.startswith("detection_results_") and file.endswith(".csv"):
            try:
                os.remove(file)
            except:
                pass

# Create the Gradio interface with Monochrome theme
with gr.Blocks(
    theme=gr.themes.Monochrome(primary_hue="pink", secondary_hue="blue"),
    title="Object Detection with YOLO",
    css="""
        .fixed-height-table {
            height: 400px !important;
            position: relative !important;
        }
        .fixed-height-table > div:nth-child(2) {
            max-height: 400px !important;
            overflow-y: auto !important;
        }
        .fixed-height-table table {
            width: 100% !important;
            border-collapse: separate !important;
            border-spacing: 0 !important;
        }
        .fixed-height-table thead {
            position: sticky !important;
            top: 0 !important;
            z-index: 2 !important;
            background: var(--background-fill-primary) !important;
        }
        .fixed-height-table th {
            background: var(--background-fill-primary) !important;
            border-bottom: 2px solid var(--border-color-primary) !important;
            padding: 8px !important;
            color: var(--body-text-color) !important;
        }
        .fixed-height-table td {
            padding: 8px !important;
        }
        /* Add gallery scroll styles */
        .gallery-scroll {
            overflow-y: auto !important;
            max-height: 500px !important;
        }
        .gallery-scroll > div {
            height: auto !important;
        }
        """
) as iface:
    gr.Markdown("# Object Detection with YOLO")
    gr.Markdown("Upload an image to detect objects using YOLO. Adjust controls to see live preview.")
    
    with gr.Row():
        # Input column
        with gr.Column(scale=1):
            input_image = gr.File(
                file_count="multiple",
                label="Input Images",
                file_types=["image"]
            )
            
            # Advanced controls
            with gr.Accordion("Advanced Controls", open=True):
                blur = gr.Slider(
                    minimum=0, maximum=10, value=0, step=0.5,
                    label="Blur Amount",
                    info="Adjust image blur (0 = no blur)"
                )
                brightness = gr.Slider(
                    minimum=0.1, maximum=2.0, value=1.0, step=0.1,
                    label="Brightness",
                    info="Adjust image brightness (1 = original)"
                )
                rotation = gr.Slider(
                    minimum=-180, maximum=180, value=0, step=5,
                    label="Rotation Angle",
                    info="Rotate image (degrees)"
                )
                confidence = gr.Slider(
                    minimum=0.0, maximum=1.0, value=0.25, step=0.01,
                    label="Confidence Threshold",
                    info="Adjust detection sensitivity"
                )
            
            detect_btn = gr.Button("Detect Objects", variant="primary")

        with gr.Column(scale=2) as output_column:
            # Preview container
            with gr.Row(visible=True) as preview_container:
                preview_image = gr.Gallery(
                    label="Live Preview",
                    show_label=True,
                    elem_id="gallery",
                    columns=2,
                    height=500,
                    allow_preview=True,
                    object_fit="contain",
                    elem_classes=["gallery-scroll"]
                )
            
            # Result container (initially hidden)
            with gr.Row(visible=False) as result_container:
                with gr.Column(scale=1):
                    final_output = gr.Gallery(
                        label="Results with Detections",
                        show_label=True,
                        elem_id="result_gallery",
                        columns=2,
                        height=500,
                        allow_preview=True,
                        object_fit="contain",
                        elem_classes=["gallery-scroll"]
                    )
                    with gr.Accordion("Detection Details", open=False):
                        detection_table = gr.Dataframe(
                            headers=["Image", "Confidence", "Top-Left", "Bottom-Right"],
                            wrap=True,
                            value=[],
                            interactive=False,
                            elem_classes=["fixed-height-table"]
                        )
                        with gr.Column():
                            export_btn = gr.Button("Export Results", variant="secondary")
                            download_file = gr.File(
                                label="Download Results",
                                show_label=False
                            )
    
    # Connect all preview events
    for component in [input_image, blur, brightness, rotation]:
        component.change(
            fn=preview_multiple,
            inputs=[input_image, blur, brightness, rotation],
            outputs=preview_image
        )

    #Clear preview container on new image uploads
    def on_new_image_upload(files, blur, brightness, rotation):
        # Get new previews
        previews = preview_multiple(files, blur, brightness, rotation)
        # Reset results container and table
        return [
            previews,                  
            gr.Row(visible=True),      
            gr.Row(visible=False),     
            None,                      
            None                       
        ]

    # Update the event connections
    input_image.change(
        fn=on_new_image_upload,
        inputs=[input_image, blur, brightness, rotation],
        outputs=[
            preview_image,
            preview_container,
            result_container,
            detection_table,
            download_file
        ]
    )

    # Component change events for live preview
    for component in [blur, brightness, rotation]:
        component.change(
            fn=preview_multiple,
            inputs=[input_image, blur, brightness, rotation],
            outputs=preview_image
        )

    # Connect the components
    def on_detect_click(*args):
        # Show "Processing" notification
        gr.Info("Detection in progress...")
        # Clear previous results
        yield [None, gr.Row(visible=False), gr.Row(visible=False), None]
        
        # Run detection
        results, data = process_multiple_images(*args)
        
        # Show completion notification
        gr.Info("Detection complete!")
        # Show new results
        yield [results, gr.Row(visible=False), gr.Row(visible=True), data]

    # Update the click handler to use streaming outputs
    detect_btn.click(
        fn=on_detect_click,
        inputs=[input_image, confidence, blur, brightness, rotation],
        outputs=[
            final_output,
            preview_container,
            result_container,
            detection_table
        ],
        queue=True  # Enable queuing for streaming
    )

    # Add export button click handler with cleanup
    def on_export_click(data):
        cleanup_temp_files()
        return export_to_csv(data)

    export_btn.click(
        fn=on_export_click,
        inputs=[detection_table],
        outputs=[download_file]
    )

iface.launch()