app.py

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

# Load the model with proper error handling
def load_model():
    model_paths = [
        'best_model.pt',
        'tree_disease_detector.pt',
        './best_model.pt',
        './tree_disease_detector.pt'
    ]
    
    # Try to load from local files first
    for path in model_paths:
        if os.path.exists(path):
            try:
                print(f"Loading model from {path}")
                model = YOLO(path)
                return model, f"Tree Disease Detection Model ({path})"
            except Exception as e:
                print(f"Error loading {path}: {e}")
                continue
    
    # Fallback to standard YOLOv8s
    try:
        print("Loading standard YOLOv8s model...")
        model = YOLO('yolov8s.pt')
        return model, "Standard YOLOv8s Model (Fallback)"
    except Exception as e:
        print(f"Error loading YOLOv8s: {e}")
        return None, "No model available"

# Load model and get status
model, model_status = load_model()

def detect_tree_disease(image, conf_threshold=0.25, iou_threshold=0.45):
    """Detect unhealthy trees in the uploaded image"""
    
    if model is None:
        return image, "Error: No model available"
    
    # Convert PIL image to numpy array
    image_np = np.array(image)
    
    # Run inference
    results = model(image_np, conf=conf_threshold, iou=iou_threshold)
    
    # Get annotated image directly from results
    annotated_img = results[0].plot()
    annotated_img = cv2.cvtColor(annotated_img, cv2.COLOR_BGR2RGB)
    annotated_img = Image.fromarray(annotated_img)
    
    # Extract detections
    detections = []
    for result in results:
        boxes = result.boxes
        if boxes is not None:
            for box in boxes:
                detection = {
                    'confidence': float(box.conf[0]),
                    'bbox': box.xyxy[0].tolist(),
                    'class': 'unhealthy'
                }
                detections.append(detection)
    
    # Create detection summary
    is_custom_model = "Tree Disease Detection Model" in model_status
    
    if is_custom_model:
        summary = f"Detected {len(detections)} unhealthy tree(s)\n\n"
        for i, det in enumerate(detections, 1):
            summary += f"Tree {i}: Confidence {det['confidence']:.2f}\n"
    else:
        summary = f"Using {model_status}\n"
        summary += f"Detected {len(detections)} object(s)\n\n"
        for i, det in enumerate(detections, 1):
            summary += f"Object {i}: Confidence {det['confidence']:.2f}\n"
    
    summary += f"\nModel Status: {model_status}"
    
    return annotated_img, summary

# Create example images (tree images)
example_images = [
    ["https://images.pexels.com/photos/1632790/pexels-photo-1632790.jpeg", 0.25, 0.45],
    ["https://images.pexels.com/photos/38537/woodland-road-falling-leaf-natural-38537.jpeg", 0.25, 0.45],
    ["https://upload.wikimedia.org/wikipedia/commons/thumb/e/eb/Ash_Tree_-_geograph.org.uk_-_590710.jpg/640px-Ash_Tree_-_geograph.org.uk_-_590710.jpg", 0.25, 0.45],
]

# Create Gradio interface
with gr.Blocks(title="Tree Disease Detection") as demo:
    gr.Markdown(f"""
    # 🌳 Tree Disease Detection with YOLOv8
    
    This model detects unhealthy/diseased trees in aerial UAV imagery. 
    Upload an image or use one of the examples below to detect diseased trees.
    
    **Current Model**: {model_status}
    """)
    
    if "Fallback" in model_status:
        gr.Markdown("""
        ⚠️ **Note**: Using a fallback model. Detection will work but won't be specific to tree diseases.
        """)
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Upload Image")
            conf_threshold = gr.Slider(
                minimum=0.0,
                maximum=1.0,
                value=0.25,
                step=0.05,
                label="Confidence Threshold"
            )
            iou_threshold = gr.Slider(
                minimum=0.0,
                maximum=1.0,
                value=0.45,
                step=0.05,
                label="IoU Threshold"
            )
            detect_button = gr.Button("Detect Tree Disease", variant="primary")
        
        with gr.Column():
            output_image = gr.Image(type="pil", label="Detection Results")
            detection_summary = gr.Textbox(label="Detection Summary", lines=10)
    
    # Set up event handler
    detect_button.click(
        fn=detect_tree_disease,
        inputs=[input_image, conf_threshold, iou_threshold],
        outputs=[output_image, detection_summary]
    )
    
    # Add examples
    gr.Examples(
        examples=example_images,
        inputs=[input_image, conf_threshold, iou_threshold],
        outputs=[output_image, detection_summary],
        fn=detect_tree_disease,
        cache_examples=False,
    )
    
    gr.Markdown("""
    ## About this Model
    
    - **Architecture**: YOLOv8s
    - **Dataset**: [PDT Dataset](https://huggingface.co/datasets/qwer0213/PDT_dataset)
    - **mAP50**: 0.933
    - **mAP50-95**: 0.659
    - **Precision**: 0.878
    - **Recall**: 0.863
    - **Classes**: 1 (unhealthy trees)
    
    ## Usage Tips
    
    - This model works best with aerial/UAV imagery
    - Optimal input resolution: 640x640 pixels
    - Adjust confidence threshold to filter detections
    - Lower IoU threshold for overlapping trees
    
    [Model Card](https://huggingface.co/IsmatS/crop_desease_detection) | 
    [Dataset](https://huggingface.co/datasets/qwer0213/PDT_dataset)
    """)

# Launch the app
if __name__ == "__main__":
    demo.launch()