app.py

import streamlit as st
from PIL import Image
from transformers import pipeline
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import xlsxwriter
import io

# Initialize session state for results, image names, and image sizes if not already present
if 'results' not in st.session_state:
    st.session_state['results'] = []
if 'image_names' not in st.session_state:
    st.session_state['image_names'] = []
if 'image_sizes' not in st.session_state:
    st.session_state['image_sizes'] = []

# Disable PyplotGlobalUseWarning
st.set_option('deprecation.showPyplotGlobalUse', False)

# Create an image classification pipeline with scores
pipe = pipeline("image-classification", model="trpakov/vit-face-expression", top_k=None)

# Streamlit app
st.title("Emotion Recognition with vit-face-expression")

# Upload images
uploaded_images = st.file_uploader("Upload images", type=["jpg", "png"], accept_multiple_files=True)

# Display thumbnail images alongside file names and sizes in the sidebar
selected_images = []
if uploaded_images:
    # Reset the image names and sizes lists each time new images are uploaded
    st.session_state['image_names'] = [img.name for img in uploaded_images]
    st.session_state['image_sizes'] = [round(img.size / 1024.0, 1) for img in uploaded_images]

    # Add a "Select All" checkbox in the sidebar
    select_all = st.sidebar.checkbox("Select All", False)
    
    for idx, img in enumerate(uploaded_images):
        image = Image.open(img)
        checkbox_key = f"{img.name}_checkbox_{idx}"  # Unique key for each checkbox
        # Display thumbnail image and checkbox in sidebar
        st.sidebar.image(image, caption=f"{img.name} {img.size / 1024.0:.1f} KB", width=40)
        selected = st.sidebar.checkbox(f"Select {img.name}", value=select_all, key=checkbox_key)         
        
        if selected:
            selected_images.append(image)

if st.button("Predict Emotions") and selected_images:
    # Predict emotion for each selected image using the pipeline
    st.session_state['results'] = [pipe(image) for image in selected_images]

# Initialize an empty DataFrame outside of the button press condition
df_emotions = pd.DataFrame()

# Generate DataFrame from results
if st.button("Generate HeatMap & DataFrame"):
    # Access the results, image names, and sizes from the session state
    results = st.session_state['results']
    image_names = st.session_state['image_names']
    image_sizes = st.session_state['image_sizes']
    if results:
        # Initialize an empty list to store all the data
        data = []

        # Iterate over the results and populate the list with dictionaries
        for i, result_set in enumerate(results):
            # Initialize a dictionary for the current set with zeros
            current_data = {
                'Happy': 0,
                'Surprise': 0,
                'Neutral': 0,
                'Sad': 0,
                'Disgust': 0,
                'Angry': 0,
                'Fear': 0,
                'Image Name': image_names[i],
                'Image Size (KB)': f"{image_sizes[i]:.1f}"  # Format the size to one decimal place
            }
            
            for result in result_set:
                # Capitalize the label and update the score in the current set
                emotion = result['label'].capitalize()
                score = round(result['score'], 4)  # Round the score to 4 decimal places
                current_data[emotion] = score
            
            # Append the current data to the data list
            data.append(current_data)

        # Convert the list of dictionaries into a pandas DataFrame
        df_emotions = pd.DataFrame(data)

        # Add a placeholder for the 'Image View' column
        df_emotions['Image View'] = [''] * len(df_emotions)
        # Add a sequence of numbers for the 'Image Num' column
        df_emotions['Image Num'] = list(range(len(df_emotions)))

        # Display the DataFrame
        st.write(df_emotions)

        # Plotting the heatmap for the first seven columns
        plt.figure(figsize=(10, 10))
        sns.heatmap(df_emotions.iloc[:, :7], annot=True, fmt=".1f", cmap='viridis')
        plt.title('Heatmap of Emotion Scores')
        plt.xlabel('Emotion Categories')
        plt.ylabel('Data Points')
        st.pyplot(plt)

        # Save the DataFrame to a CSV file without the 'Image View' and 'Image Num' columns
        df_emotions.drop(columns=['Image View', 'Image Num']).to_csv('emotion_scores.csv', index=False)
        st.success('DataFrame generated and saved as emotion_scores.csv')

        with open('emotion_scores.csv', 'r') as f:
            csv_file = f.read()

        st.download_button(
            label='Download Emotion Scores as CSV',
            data=csv_file,
            file_name='emotion_scores.csv',
            mime='text/csv',
        )

        # Create a BytesIO buffer for the Excel file
        output = io.BytesIO()

        # Create a new Excel writer object using the buffer as the file
        writer = pd.ExcelWriter(output, engine='xlsxwriter')
        df_emotions.to_excel(writer, index=False, header=True)

        # Access the xlsxwriter workbook and worksheet objects
        workbook  = writer.book
        worksheet = writer.sheets['Sheet1']

        # Set the column width and row height
        worksheet.set_column('A:G', 8)  # Set width for columns A-G
        worksheet.set_column('H:H', 22)  # Set width for column H (Image Name)   
        worksheet.set_column('I:I', 14)  # Set width for column I (Image Size)   
        worksheet.set_column('J:J', 12)  # Set width for column J (Image View)   
        worksheet.set_column('K:K', 12)  # Set width for column K (Image Num)

        
        for row_num in range(len(df_emotions) + 1):  # +1 to include the header row
            worksheet.set_row(row_num, 52)  # Set the row height to 38

        # Iterate over the images and insert them into the 'Image View' column
        for idx, image in enumerate(selected_images):
            # Convert the image to a format that can be inserted into Excel
            image_stream = io.BytesIO()
            image.save(image_stream, format='JPEG')  # image.save(image_stream, format='JPEG')  # Save the image as JPEG; or PNG
            image_stream.seek(0)
            # Calculate the scaling factor to fit the image inside the cell
            cell_width = 64
            scale_factor = cell_width / image.width
            # Insert the image into the cell
            worksheet.insert_image(f'J{idx + 2}', 'image.jpg', {      #or image.png
                'image_data': image_stream,
                'x_scale': scale_factor,
                'y_scale': scale_factor,
                'x_offset': 2,
                'y_offset': 2,
                'positioning': 1
            })

        # Close the writer object
        writer.close()

        # Rewind the buffer
        output.seek(0)

        # Use Streamlit's download button to offer the Excel file for download
        st.download_button(
            label='Download Emotion Scores as Excel',
            data=output,
            file_name='emotion_scores.xlsx',
            mime='application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
        )