app.py

import gradio as gr
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import keras_cv
import keras

# COCO class labels (80 classes)
COCO_CLASSES = [
    "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train",
    "truck", "boat", "traffic light", "fire hydrant", "stop sign",
    "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep",
    "cow", "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella",
    "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard",
    "sports ball", "kite", "baseball bat", "baseball glove", "skateboard",
    "surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork",
    "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange",
    "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair",
    "couch", "potted plant", "bed", "dining table", "toilet", "tv",
    "laptop", "mouse", "remote", "keyboard", "cell phone", "microwave",
    "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase",
    "scissors", "teddy bear", "hair drier", "toothbrush",
]

# Color palette for bounding boxes
COLORS = [
    "#FF6B6B", "#4ECDC4", "#45B7D1", "#96CEB4", "#FFEAA7",
    "#DDA0DD", "#98D8C8", "#F7DC6F", "#BB8FCE", "#85C1E9",
    "#F8C471", "#82E0AA", "#F1948A", "#AED6F1", "#D7BDE2",
]


def load_model():
    """Load pretrained YOLOv8 model from KerasCV."""
    model = keras_cv.models.YOLOV8Detector.from_preset(
        "yolo_v8_m_pascalvoc",
        bounding_box_format="xyxy",
    )
    return model


print("Loading model...")
model = load_model()
print("Model loaded!")


def detect_objects(image, confidence_threshold=0.5):
    """Run object detection on a single image."""
    if image is None:
        return None

    orig_image = Image.fromarray(image)
    orig_w, orig_h = orig_image.size

    # Resize for model input
    input_size = 640
    resized = orig_image.resize((input_size, input_size))
    img_array = np.array(resized, dtype="float32")
    input_batch = np.expand_dims(img_array, axis=0)

    # Run prediction
    predictions = model.predict(input_batch)

    boxes = predictions["boxes"][0]
    classes = predictions["classes"][0]
    confidence = predictions["confidence"][0]

    # Convert to numpy if needed
    if hasattr(boxes, "numpy"):
        boxes = boxes.numpy()
        classes = classes.numpy()
        confidence = confidence.numpy()

    # Draw results on original image
    draw = ImageDraw.Draw(orig_image)

    try:
        font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 16)
        small_font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 13)
    except OSError:
        font = ImageFont.load_default()
        small_font = font

    detections_found = 0

    for i in range(len(boxes)):
        score = float(confidence[i])
        if score < confidence_threshold:
            continue

        cls_id = int(classes[i])
        if cls_id < 0 or cls_id >= len(COCO_CLASSES):
            label = f"class_{cls_id}"
        else:
            label = COCO_CLASSES[cls_id]

        # Scale boxes from resized coords back to original image
        x1 = float(boxes[i][0]) * orig_w / input_size
        y1 = float(boxes[i][1]) * orig_h / input_size
        x2 = float(boxes[i][2]) * orig_w / input_size
        y2 = float(boxes[i][3]) * orig_h / input_size

        color = COLORS[cls_id % len(COLORS)]

        # Draw bounding box
        draw.rectangle([x1, y1, x2, y2], outline=color, width=3)

        # Draw label background + text
        text = f"{label} {score:.0%}"
        bbox = draw.textbbox((x1, y1), text, font=font)
        text_w = bbox[2] - bbox[0]
        text_h = bbox[3] - bbox[1]
        draw.rectangle([x1, y1 - text_h - 6, x1 + text_w + 8, y1], fill=color)
        draw.text((x1 + 4, y1 - text_h - 4), text, fill="white", font=font)

        detections_found += 1

    status = f"Found {detections_found} object(s)" if detections_found else "No objects detected"
    return orig_image, status


# Build the Gradio interface
with gr.Blocks(title="Keras Object Detection") as demo:
    gr.Markdown("# Object Detection with KerasCV YOLOv8")
    gr.Markdown("Upload an image to detect objects using a pretrained YOLOv8 model.")

    with gr.Row():
        with gr.Column():
            input_image = gr.Image(label="Upload Image", type="numpy")
            threshold = gr.Slider(
                minimum=0.1,
                maximum=0.95,
                value=0.5,
                step=0.05,
                label="Confidence Threshold",
            )
            run_btn = gr.Button("Detect Objects", variant="primary")
        with gr.Column():
            output_image = gr.Image(label="Detections")
            status_text = gr.Textbox(label="Status", interactive=False)

    run_btn.click(
        fn=detect_objects,
        inputs=[input_image, threshold],
        outputs=[output_image, status_text],
    )

demo.launch()