app.py

import os
import cv2
import uuid

import gradio as gr
import numpy as np

import neovision

import random

MARKDOWN = """
# Welcome to VisionB 🧠 + 📸

Meet VisionB, your Visual Agent that combines the power of advanced GPT models with real-time visual inputs. Engage in interactive dialogues, ask questions, and gain insights with the added context of images from your webcam. Experience a new dimension of interaction where vision and conversational AI meet.
"""

connector = neovision.OpanAIConnector()


def generate_liveness_challenge(image_details):
    # Based on the image details, generate a challenge
    challenges = []

    if 'glasses' in image_details:
        challenges.append("Please take off your glasses and hold them in your hand.")
    if 'smiling' in image_details:
        challenges.append("Please take another picture with a neutral expression.")
    
    # You can add more contextual clues and corresponding challenges

    # Generic challenges if no specific detail is detected
    if not challenges:
        challenges = [
            "Please hold up 5 fingers.",
            "Use your hand to cover one of your eyes.",
            "Make an OK sign with your hand and hold it up to your chin."
        ]

    return random.choice(challenges)

def save_image_to_drive(image: np.ndarray) -> str:
    image_filename = f"{uuid.uuid4()}.jpeg"
    image_directory = "data"
    os.makedirs(image_directory, exist_ok=True)
    image_path = os.path.join(image_directory, image_filename)
    cv2.imwrite(image_path, image)
    return image_path

def respond(image: np.ndarray, prompt: str, chat_history=None):

    # Initialize chat_history as an empty list if it's None
    if chat_history is None:
        chat_history = []

    image = np.fliplr(image)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image_path = save_image_to_drive(image)


    # response = connector.simple_prompt(image=image, prompt=prompt)

    # If the user's prompt is 'verify', we start the liveness challenge
    if 'verify' in prompt.lower():
        # Use the image details to generate a challenge
        # This is where you'd use the AI's analysis of the image to tailor the challenge
        # For simplicity, the details are hard-coded here
        # image_details = "A Person Wearing glasses"  # Placeholder for actual analysis

        # Get the image details from the AI model
        image_details = connector.simple_prompt(image=image, prompt="What details can you describe from this image?")

        # print(ai_response)

        challenge = generate_liveness_challenge(image_details)
        response = f"For liveness verification, {challenge}"
    else:
        # For any other prompt, just process normally
        response = connector.simple_prompt(image=image, prompt=prompt)


    chat_history.append(((image_path,), None))
    chat_history.append((prompt, response))
    return "", chat_history


with gr.Blocks() as demo:
    gr.Markdown(MARKDOWN)
    with gr.Row():
        webcam = gr.Image(sources=["webcam"], streaming=True)
        with gr.Column():
            chatbot = gr.Chatbot(height=500)
            message = gr.Textbox()
            clear_button = gr.ClearButton([message, chatbot])

    message.submit(respond, [webcam, message, chatbot], [message, chatbot])

demo.launch(debug=False, show_error=True)