import gradio as gr
from PIL import Image, ImageDraw, ImageFont
import scipy.io.wavfile as wavfile


# Use a pipeline as a high-level helper
from transformers import pipeline

# model_path = ("../Models/models--facebook--detr-resnet-50/snapshots"
#               "/1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b")

# object_detector = pipeline("object-detection",
#                 model=model_path)

object_detector = pipeline("object-detection",
                model="facebook/detr-resnet-50")

narrator = pipeline("text-to-speech",
                    model="kakao-enterprise/vits-ljs")


# Define the function to generate audio from text
def generate_audio(text):
    # Generate the narrated text
    narrated_text = narrator(text)

    # Save the audio to a WAV file
    wavfile.write("output.wav", rate=narrated_text["sampling_rate"],
                  data=narrated_text["audio"][0])

    # Return the path to the saved audio file
    return "output.wav"



def read_objects(detection_objects):
    # Initialize counters for each object label
    object_counts = {}

    # Count the occurrences of each label
    for detection in detection_objects:
        label = detection['label']
        if label in object_counts:
            object_counts[label] += 1
        else:
            object_counts[label] = 1

    # Generate the response string
    response = "This picture contains"
    labels = list(object_counts.keys())
    for i, label in enumerate(labels):
        response += f" {object_counts[label]} {label}"
        if object_counts[label] > 1:
            response += "s"
        if i < len(labels) - 2:
            response += ","
        elif i == len(labels) - 2:
            response += " and"

    response += "."

    return response



def draw_bounding_boxes(image, detections, font_path=None, font_size=50):
    # Make a copy of the image to draw on
    draw_image = image.copy()
    draw = ImageDraw.Draw(draw_image)

    # Load custom font or default font if path not provided
    if font_path:
        font = ImageFont.truetype(font_path, font_size)
    else:
        # When font_path is not provided, load default font but its size is fixed
        font = ImageFont.load_default()
        # Increase font size workaround by using a TTF font file, if needed, can download and specify the path

    for detection in detections:
        box = detection['box']
        xmin = box['xmin']
        ymin = box['ymin']
        xmax = box['xmax']
        ymax = box['ymax']

        # Draw the bounding box
        draw.rectangle([(xmin, ymin), (xmax, ymax)], outline="red", width=5)

        # Optionally, you can also draw the label and score
        label = detection['label']
        score = detection['score']
        text = f"{label} {score:.2f}"

        # Draw text with background rectangle for visibility
        if font_path:  # Use the custom font with increased size
            text_size = draw.textbbox((xmin, ymin), text, font=font)
        else:
            # Calculate text size using the default font
            text_size = draw.textbbox((xmin, ymin), text)

        draw.rectangle([(text_size[0], text_size[1]), (text_size[2], text_size[3])], fill="red")
        draw.text((xmin, ymin), text, fill="white", font=font)

    return draw_image



def detect_object(image):
    raw_image = image
    output = object_detector(raw_image)
    processed_image = draw_bounding_boxes(raw_image, output)
    natural_text = read_objects(output)
    processed_audio = generate_audio(natural_text)
    return processed_image, processed_audio
    

examples = [
    ["example1.jpg"],
    ["example2.jpg"],
]    

demo = gr.Interface(fn=detect_object,
                    inputs=[gr.Image(label="Select Image",type="pil")],
                    theme='freddyaboulton/dracula_revamped',
                    outputs=[gr.Image(label="Processed Image", type="pil")],
                    examples = examples,
                    title="Object Detector",
                    description="Detect objects in the input image with bounding boxes with audio description.")
demo.launch()