app.py

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 = ("../Model/models--facebook--detr-resnet-50/snapshots"
              "/1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b")

tts_model_path = ("../Model/models--kakao-enterprise--vits-ljs/snapshots"
                  "/3bcb8321394f671bd948ebf0d086d694dda95464")


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

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


# object_detector = pipeline("object-detection", model=model_path)
# narrator = pipeline("text-to-speech", model=tts_model_path)


def generate_audio(text):
    narrated_text = narrator(text)
    wavfile.write("finetuned_output.wav",
                  rate=narrated_text["sampling_rate"],
                  data=narrated_text["audio"][0])
    return "finetuned_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, detection_results):
    """
    Draws bounding boxes on the provided image based on the detection results.

    Parameters:
        image (PIL.Image): The input image to be annotated.
        detection_results (list): A list of dictionaries, each containing the detected object details.

    Returns:
        PIL.Image: The image with bounding boxes drawn around the detected objects.
    """
    # Convert the input image to ImageDraw object to draw on it
    draw = ImageDraw.Draw(image)

    # Iterate through each detection result
    for result in detection_results:
        # Extract the bounding box coordinates and label
        box = result['box']
        label = result['label']
        score = result['score']

        # Define coordinates for the bounding box
        xmin, ymin, xmax, ymax = box['xmin'], box['ymin'], box['xmax'], box['ymax']

        # Draw the bounding box (with a red outline)
        draw.rectangle([xmin, ymin, xmax, ymax], outline="red", width=3)

        # Optionally, add label with score near the bounding box
        text = f"{label} ({score * 100:.1f}%)"
        draw.text((xmin, ymin - 10), text, fill="red")

    return image

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



demo = gr.Interface(fn = detect_objects,
                    inputs=[gr.Image(label="Select Image",type="pil")],
                    outputs=[gr.Image(label="Summarized Text ",type="pil"), gr.Audio(label="Generated Audio")],
                    title="@SherryAhuja Project : Object Detection with Audio",
                    description="This AI application will be used to Detect objects in an image and generate audio.",)
demo.launch()