app.py

import streamlit as st
import torch
import bitsandbytes
import accelerate
import scipy
import copy
from PIL import Image
import torch.nn as nn
from my_model.object_detection import detect_and_draw_objects
from my_model.captioner.image_captioning import get_caption
from my_model.gen_utilities import free_gpu_resources
from my_model.KBVQA import KBVQA, prepare_kbvqa_model



def answer_question(caption, detected_objects_str, question, model):

    answer = model.generate_answer(question, caption, detected_objects_str)
    return answer

def get_caption(image):
    return "Generated caption for the image"

def free_gpu_resources():
    pass

# Sample images (assuming these are paths to your sample images)
sample_images = ["Files/sample1.jpg", "Files/sample2.jpg", "Files/sample3.jpg", 
                 "Files/sample4.jpg", "Files/sample5.jpg", "Files/sample6.jpg", 
                 "Files/sample7.jpg"]



def analyze_image(image, model):
    st.write("Analyzing . . .")
    caption = model.get_caption(image)
    image_with_boxes, detected_objects_str = model.detect_objects(image)
    return caption, detected_objects_str
    

def image_qa_app(kbvqa):
    if 'images_data' not in st.session_state:
        st.session_state['images_data'] = {}

    # Display sample images as clickable thumbnails
    st.write("Choose from sample images:")
    cols = st.columns(len(sample_images))
    for idx, sample_image_path in enumerate(sample_images):
        with cols[idx]:
            image = Image.open(sample_image_path)
            st.image(image, use_column_width=True)
            if st.button(f'Select Sample Image {idx + 1}', key=f'sample_{idx}'):
                process_new_image(sample_image_path, image, kbvqa)

    # Image uploader
    uploaded_image = st.file_uploader("Or upload an Image", type=["png", "jpg", "jpeg"])
    if uploaded_image is not None:
        process_new_image(uploaded_image.name, Image.open(uploaded_image), kbvqa)

    # Display and interact with each uploaded/selected image
    for image_key, image_data in st.session_state['images_data'].items():
        st.image(image_data['image'], caption=f'Uploaded Image: {image_key[-11:]}', use_column_width=True)
        if not image_data['analysis_done']:
            if st.button('Analyze Image', key=f'analyze_{image_key}'):
                caption, detected_objects_str = analyze_image(image_data['image'], kbvqa)
                image_data['caption'] = caption
                image_data['detected_objects_str'] = detected_objects_str
                image_data['analysis_done'] = True

        if image_data['analysis_done']:
            question = st.text_input(f"Ask a question about this image ({image_key}):", key=f'question_{image_key}')
            if st.button('Get Answer', key=f'answer_{image_key}'):
                answer = answer_question(image_data['caption'], image_data['detected_objects_str'], question, kbvqa)
                image_data['qa_history'].append((question, answer))

            for q, a in image_data['qa_history']:
                st.text(f"Q: {q}\nA: {a}\n")

def process_new_image(image_key, image, kbvqa):
    """Process a new image and update the session state."""
    if image_key not in st.session_state['images_data']:
        st.session_state['images_data'][image_key] = {
            'image': image,
            'caption': '',
            'detected_objects_str': '',
            'qa_history': [],
            'analysis_done': False
        }

def run_inference():
    st.title("Run Inference")

    method = st.selectbox("Choose a method:", ["Fine-Tuned Model", "In-Context Learning (n-shots)"], index=0)
    detection_model = st.selectbox("Choose a model for object detection:", ["yolov5", "detic"], index=0)
    confidence_level = st.slider("Select minimum detection confidence level", min_value=0.1, max_value=0.9, value=0.2 if detection_model == "yolov5" else 0.4, step=0.1)

    # Check for changes in model or confidence level
    model_changed = (st.session_state.get('detection_model') != detection_model)
    confidence_changed = (st.session_state.get('confidence_level') != confidence_level)

    if model_changed or confidence_changed:
        st.session_state['detection_model'] = detection_model
        st.session_state['confidence_level'] = confidence_level
        st.warning("Detection model or confidence level changed. Please reload the model, this will take few seconds :)")


    
    # Initialize session state for the model

    if method == "Fine-Tuned Model":
        if 'kbvqa' not in st.session_state:
            st.session_state['kbvqa'] = None
    
        # Button to load KBVQA models



        if st.button('Load Model'):
            if st.session_state.get('kbvqa') and not model_changed and not confidence_changed:
                st.write("Model already loaded.")
            else:
                st.text("Loading the model will take no more than a few minutes . .")
                st.session_state['kbvqa'] = prepare_kbvqa_model(detection_model)
                st.session_state['kbvqa'].detection_confidence = confidence_level
                st.success("Model loaded with updated settings.")

        if st.session_state.get('kbvqa'):
            st.write("Model is ready for inference.")
            image_qa_app(st.session_state['kbvqa'])

    else: 
        st.write('Model is not ready for inference yet')
        # here goes the code for n-shot learning

            
# Main function
def main():

        
    st.sidebar.title("Navigation")
    selection = st.sidebar.radio("Go to", ["Home", "Dataset Analysis", "Evaluation Results", "Finetuning and Evaluation Results", "Run Inference", "Dissertation Report"])
    st.sidebar.write("More Pages will follow .. ")

    if selection == "Home":
        st.title("MultiModal Learning for Knowledg-Based Visual Question Answering")
        st.write("Home page content goes here...")
        
    elif selection == "Dissertation Report":
        st.title("Dissertation Report")
        st.write("Click the link below to view the PDF.")
        # Example to display a link to a PDF
        st.download_button(
            label="Download PDF",
            data=open("Files/Dissertation Report.pdf", "rb"),
            file_name="example.pdf",
            mime="application/octet-stream"
        )

        
    elif selection == "Evaluation Results":
        st.title("Evaluation Results")
        st.write("This is a Place Holder until the contents are uploaded.")

        
    elif selection == "Dataset Analysis":
        st.title("OK-VQA Dataset Analysis")
        st.write("This is a Place Holder until the contents are uploaded.")

    elif selection == "Finetuning and Evaluation Results":
        st.title("Finetuning and Evaluation Results")
        st.write("This is a Place Holder until the contents are uploaded.")


    elif selection == "Run Inference":
        run_inference()

    elif selection == "More Pages will follow .. ":
        st.title("Staye Tuned")
        st.write("This is a Place Holder until the contents are uploaded.")
            


if __name__ == "__main__":
    main()