app.py

#!/usr/bin/env python
# coding: utf-8

# In[5]:


import streamlit as st
from PIL import Image
import torch
import torch
import requests
from transformers import BlipProcessor, BlipForQuestionAnswering,BlipImageProcessor, AutoProcessor
from transformers import BlipConfig
from datasets import load_dataset
from torch.utils.data import DataLoader
from tqdm.notebook import tqdm

import numpy as np
import matplotlib.pyplot as plt
from IPython.display import display

text_processor = BlipProcessor.from_pretrained("Salesforce/blip-vqa-base")
image_processor = BlipImageProcessor.from_pretrained("Salesforce/blip-vqa-base")
model = BlipForQuestionAnswering.from_pretrained(r"blip_model_v2_epo89" )


def preprocess_image(image):
    # Your image preprocessing logic here...
    # Example: Resize image to 128x128 pixels
    image = image.resize((128, 128))
    image_encoding = image_processor(image,
                                     do_resize=True,
                                     size=(128, 128),
                                     return_tensors="pt")
    return image_encoding["pixel_values"][0]

def preprocess_text(text, max_length=32):
    # Your text preprocessing logic here...
    encoding = text_processor(
        None,
        text,
        padding="max_length",
        truncation=True,
        max_length=max_length,
        return_tensors="pt"
    )
    
    for k, v in encoding.items():
        encoding[k] = v.squeeze()
    return encoding

def predict(image, question):
    # Preprocess image
    pixel_values = preprocess_image(image).unsqueeze(0)

    # Preprocess text
    encoding = preprocess_text(question)

    # Print shapes for debugging
    #print("Pixel Values Shape:", pixel_values.shape)
    #print("Input IDs Shape:", encoding['input_ids'].unsqueeze(0).shape)

    # Perform prediction using your model
    # Example: Replace this with your actual prediction logic
    model.eval()
    outputs = model.generate(pixel_values=pixel_values, input_ids=encoding['input_ids'].unsqueeze(0))

    prediction_result = text_processor.decode(outputs[0], skip_special_tokens=True)

    return prediction_result

def main():
    # Set page title and configure page layout
    st.set_page_config(
        page_title="PathoAgent",
        page_icon=":microscope:",
        layout="wide"
    )

    # Add header with styled text
    st.title(":microscope: PathoAgent")
    st.markdown(
        """
        <style>
            body {
                background-color: #f1f1f1;
            }
            .header {
                text-align: center;
                padding: 20px;
                background-color: #3498db;
            }
            .subheader {
                color: #fff;
                text-align: center;
                padding-bottom: 20px;
            }
        </style>
        """,
        unsafe_allow_html=True
    )
    
    st.markdown("<div class='header'><h3 class='subheader'>Medical Image Analysis for Pathology</h3></div>", unsafe_allow_html=True)
    st.markdown("<hr style='border: 1px solid #ddd;'>", unsafe_allow_html=True)

    # Navigation bar
    nav_option = st.sidebar.radio("Navigation", ["Home", "Sample Images", "Upload Image"])
    
    if nav_option == "Home":
        home()
    elif nav_option == "Sample Images":
        sample_images()
    elif nav_option == "Upload Image":
        upload_image()

def home():
    st.header("Welcome to PathoAgent!")
    st.write(
        "PathoAgent is an AI-powered medical image analysis tool designed for pathology diagnostics. "
        "It empowers healthcare professionals with accurate predictions and insights from medical images. "
        "Choose an option from the sidebar to get started."
    )
    
    st.header("About PathoAgent")
    st.write(
        "PathoAgent leverages advanced VQA algorithms to analyze medical images related to pathology. "
        "Whether you want to upload your own images or use our sample images, PathoAgent provides predictions for pathology-related questions. "
        "Explore the features and capabilities to enhance your diagnostic process."
    )

def sample_images():
    st.header("Sample Images")

    # Sample images
    example_image = {
        "Sample 1": "img_0002.jpg",
    }

    # Button to load sample images
    if st.button("Load Example Images"):
        
        sample_image = Image.open(example_image).convert('RGB')
        st.image(sample_image, caption=f"Example Image", use_column_width=True)

        # Text input for each sample image
        text_input = st.text_area(f"Input Question:")

        # Predict button for each sample image
        if st.button(f"Predict"):
            if text_input:
                # Perform prediction
                prediction_result = predict(sample_image, text_input)

                # Display input text
                st.subheader(f"Input Question:")
                st.write(text_input)

                # Display prediction result
                st.subheader(f"Prediction Result:")
                st.write(prediction_result)

def upload_image():
    st.header("Upload Image")

    # Image upload
    uploaded_file = st.file_uploader("Choose a file", type=["jpg", "png", "jpeg"])

    # Text input
    st.subheader("Input Question")
    text_input = st.text_area("Enter text here:")

    # Display uploaded image
    if uploaded_file is not None:
        image = Image.open(uploaded_file).convert('RGB')
        st.image(image, caption="Uploaded Image.", use_column_width=True)

    # Predict button
    if st.button("Predict"):
        if uploaded_file is not None and text_input:
            # Perform prediction
            prediction_result = predict(image, text_input)

            # Display input text
            st.subheader("Input Question:")
            st.write(text_input)

            # Display prediction result
            st.subheader("Prediction Result:")
            st.write(prediction_result)

if __name__ == "__main__":
    main()