app.py

import streamlit as st
import os
import json
import glob
from typing import List, Dict, Tuple, Optional
import tempfile
from PIL import Image, ImageDraw, ImageFont
import imageio
import streamlit.components.v1 as components
import base64

st.set_page_config(
    page_title="Computer Use Interaction Visualizer", 
    page_icon="🖥️",
    layout="wide"
)

# Constants
DATA_ROOT = "website_data_v2"
ACTION_PAIRS_TO_SKIP = set(["CURSOR_POSITIONING", "KEYBOARD_INPUT_MOVE_START"])

def scan_data_directory() -> Dict:
    """
    Scans the data directory structure and returns a dictionary
    representing the hierarchy of categories, websites, and sessions.
    """
    data_structure = {}
    
    # Walk through all categories
    for category in os.listdir(DATA_ROOT):
        category_path = os.path.join(DATA_ROOT, category)
        if not os.path.isdir(category_path):
            continue
            
        data_structure[category] = {}
        
        # Walk through all websites in this category
        for website in os.listdir(category_path):
            website_path = os.path.join(category_path, website)
            if not os.path.isdir(website_path):
                continue
                
            data_structure[category][website] = []
            
            # Walk through all sessions for this website
            for session in os.listdir(website_path):
                session_path = os.path.join(website_path, session)
                if not os.path.isdir(session_path):
                    continue
                    
                # Check if the session folder contains a nested folder with the same name
                nested_session_path = os.path.join(session_path, session)
                if os.path.isdir(nested_session_path):
                    # Check if the nested folder contains a screenshots directory
                    screenshots_path = os.path.join(nested_session_path, "screenshots")
                    if os.path.isdir(screenshots_path):
                        data_structure[category][website].append(session)
    
    return data_structure

def load_session_data(category: str, website: str, session: str) -> Tuple[Dict, List[Dict], List[str]]:
    """
    Loads all the data for a specific session including metadata, actions, and screenshots.
    """
    # Paths to data files
    session_path = os.path.join(DATA_ROOT, category, website, session)
    metadata_path = os.path.join(session_path, "metadata.json")
    nested_session_path = os.path.join(session_path, session)
    action_log_path = os.path.join(nested_session_path, "action_log.json")
    screenshots_path = os.path.join(nested_session_path, "screenshots")
    
    # Load metadata
    with open(metadata_path, 'r') as f:
        metadata = json.load(f)
    
    # Load action log
    with open(action_log_path, 'r') as f:
        action_data = json.load(f)
    
    # Get sorted list of screenshots
    screenshots = sorted(glob.glob(os.path.join(screenshots_path, "frame_*.png")))
    screenshots = [os.path.basename(s) for s in screenshots]
    
    return metadata, action_data, screenshots

def group_actions_by_type(actions: List[Dict]) -> Dict[str, List[Dict]]:
    """
    Groups actions by their type for easier analysis.
    """
    grouped = {}
    for action in actions:
        action_type = action.get("type", "UNKNOWN")
        if action_type not in grouped:
            grouped[action_type] = []
        grouped[action_type].append(action)
    return grouped

def find_before_after_pairs(actions: List[Dict]) -> List[Tuple[Optional[Dict], Optional[Dict]]]:
    """
    Analyzes the action log to find before/after pairs of actions.
    Returns a list of (before, after) action tuples.
    """
    pairs = []
    i = 0
    
    while i < len(actions):
        before_action = None
        after_action = None
        
        # Check if current action is a "START" or "BEFORE" action
        if "_START" in actions[i]["type"] or "BEFORE" in actions[i]["description"]:
            before_action = actions[i]
            
            # Look for the corresponding "COMPLETE" or "AFTER" action
            j = i + 1
            while j < len(actions) and j < i + 5:  # Look ahead up to 5 actions
                if ("_COMPLETE" in actions[j]["type"] or "AFTER" in actions[j]["description"]) and \
                   actions[i]["type"].split("_START")[0] in actions[j]["type"]:
                    after_action = actions[j]
                    i = j  # Skip ahead to after the COMPLETE action
                    break
                j += 1
                
        # If we couldn't find a matching after action, just use this action alone
        if not after_action and "_COMPLETE" not in actions[i]["type"]:
            after_action = None
            before_action = actions[i]
        
        if before_action:
            pairs.append((before_action, after_action))
        
        i += 1
    
    return pairs

def extract_frame_number(screenshot_name: str) -> int:
    """Extract the frame number from a screenshot filename."""
    try:
        return int(screenshot_name.split('_')[1].split('.')[0])
    except (IndexError, ValueError):
        return -1

def draw_cursor(image, mouse_pos):
    """
    Draw a mouse cursor on the image at the specified position.
    Args:
        image (PIL.Image): The image to draw on
        mouse_pos (tuple): The (x, y) position of the mouse cursor
    """
    if not mouse_pos:
        return image
    
    draw = ImageDraw.Draw(image)
    
    mouse_x, mouse_y = mouse_pos
    
    # Draw a blue circle at the mouse position (similar to generate_trajectory.py)
    circle_radius = 15
    circle_outline = 3
    
    # Draw outer blue circle
    draw.ellipse(
        [(mouse_x - circle_radius, mouse_y - circle_radius), 
         (mouse_x + circle_radius, mouse_y + circle_radius)],
        outline=(0, 0, 255, 255),
        width=circle_outline
    )
    
    # Draw smaller inner circle for better visibility
    draw.ellipse(
        [(mouse_x - 5, mouse_y - 5), 
         (mouse_x + 5, mouse_y + 5)],
        fill=(0, 0, 255, 128)  # Semi-transparent blue
    )
    
    return image

def create_transition_gif(before_path, after_path, output_path, duration=10.0):
    """
    Create a GIF animation transitioning between before and after screenshots.
    
    Args:
        before_path (str): Path to the before screenshot
        after_path (str): Path to the after screenshot
        output_path (str): Path to save the resulting GIF
        duration (float): Duration of each frame in seconds
    """
    try:
        # Load the images
        images = []
        before_img = Image.open(before_path)
        after_img = Image.open(after_path)
        
        # Ensure both images are the same size (resize if needed)
        if before_img.size != after_img.size:
            after_img = after_img.resize(before_img.size)
        
        # Add text overlay to images
        before_img_with_text = add_header_text(before_img.copy(), "BEFORE")
        after_img_with_text = add_header_text(after_img.copy(), "AFTER")
        
        # Add the before image 60 times
        for _ in range(60):
            images.append(before_img_with_text)
        
        # Create a blank transition frame (copy of before image with reduced opacity)
        transition_img = Image.new('RGBA', before_img.size, (0, 0, 0, 0))
        
        # Add the transition image 60 times
        for _ in range(60):
            images.append(transition_img)
        
        # Add the after image 60 times
        for _ in range(60):
            images.append(after_img_with_text)
        
        # Save as GIF with default duration
        imageio.mimsave(output_path, images, loop=0, duration=duration)
        
        return output_path
    except Exception as e:
        st.error(f"Error creating GIF: {e}")
        return None

def add_header_text(image, text):
    """
    Add large header text to the top of an image.
    
    Args:
        image (PIL.Image): The image to add text to
        text (str): The text to add
    
    Returns:
        PIL.Image: Image with text overlay
    """
    # Convert to RGBA if needed to support transparency
    if image.mode != 'RGBA':
        image = image.convert('RGBA')
    
    # Create a drawing context
    draw = ImageDraw.Draw(image)
    
    # Find a suitable font size based on image width - increased multiplier for larger font
    font_size = int(image.width * 0.5)
    font = ImageFont.load_default()

    # Get text dimensions using the selected font
    try:
        text_bbox = draw.textbbox((0, 0), text, font=font)
        text_width = text_bbox[2] - text_bbox[0]
        text_height = text_bbox[3] - text_bbox[1]
    except AttributeError: # Handle cases where font might be the basic default font without bbox support
        # Estimate text size for basic default font (less accurate)
        text_width = len(text) * font_size * 0.6 # Rough estimation
        text_height = font_size # Rough estimation
        st.warning("Could not accurately determine text dimensions with the fallback font.")


    # Position text at the top center of the image with some padding
    # Ensure position calculation doesn't result in negative coordinates
    x_pos = max(0, (image.width - text_width) // 2)
    y_pos = 30 # Keep top padding consistent
    position = (x_pos, y_pos)
    
    # Define padding around the text for the background
    padding_x = 30
    padding_y = 15

    # Draw semi-transparent background fitted around the text with padding
    background_bbox = (
        max(0, position[0] - padding_x), # Ensure background doesn't go off-left
        max(0, position[1] - padding_y), # Ensure background doesn't go off-top
        min(image.width, position[0] + text_width + padding_x), # Ensure background doesn't go off-right
        min(image.height, position[1] + text_height + padding_y) # Ensure background doesn't go off-bottom
    )
    # Only draw background if dimensions are valid
    if background_bbox[2] > background_bbox[0] and background_bbox[3] > background_bbox[1]:
        draw.rectangle(background_bbox, fill=(0, 0, 0, 180)) # Semi-transparent black background
    
    # Draw text in white
    draw.text(position, text, font=font, fill=(255, 255, 255, 255)) # White text
    
    return image

def create_cursor_legend() -> Image.Image:
    """
    Creates a small image explaining what the mouse cursor looks like in the visualizations.
    """
    # Create a small transparent image for the legend
    legend_img = Image.new('RGBA', (180, 80), (255, 255, 255, 220))
    draw = ImageDraw.Draw(legend_img)
    
    # Draw the cursor at the center-left of the image
    cursor_x, cursor_y = 40, 40
    
    # Draw outer blue circle
    circle_radius = 15
    circle_outline = 3
    draw.ellipse(
        [(cursor_x - circle_radius, cursor_y - circle_radius), 
         (cursor_x + circle_radius, cursor_y + circle_radius)],
        outline=(0, 0, 255, 255),
        width=circle_outline
    )
    
    # Draw smaller inner circle for better visibility
    draw.ellipse(
        [(cursor_x - 5, cursor_y - 5), 
         (cursor_x + 5, cursor_y + 5)],
        fill=(0, 0, 255, 128)  # Semi-transparent blue
    )
    
    # Add text explanation
    font = ImageFont.load_default()
    draw.text((cursor_x + 30, cursor_y - 15), "Mouse Cursor", font=font, fill=(0, 0, 0, 255))
    draw.text((cursor_x + 30, cursor_y + 5), "Position", font=font, fill=(0, 0, 0, 255))
    
    # Draw border around legend
    draw.rectangle([(0, 0), (179, 79)], outline=(200, 200, 200, 255), width=1)
    
    return legend_img

def main():
    st.title("Computer Use Interaction Visualizer")
    
    # Scan data directory
    data_structure = scan_data_directory()
    
    if not data_structure:
        st.error(f"No data found in {DATA_ROOT}. Please check the directory path.")
        return
    
    # Sidebar for navigation
    st.sidebar.title("Navigation")
    
    # Reset navigation function
    def reset_navigation():
        st.session_state.current_pair_index = 0
    
    # Select category
    category_options = list(data_structure.keys())
    category = st.sidebar.selectbox("Select Category", category_options, on_change=reset_navigation)
    
    if not category:
        st.info("Please select a category.")
        return
    
    # Select website
    website_options = list(data_structure[category].keys())
    website = st.sidebar.selectbox("Select Website", website_options, on_change=reset_navigation)
    
    if not website:
        st.info("Please select a website.")
        return
    
    # Select session
    session_options = data_structure[category][website]
    session = st.sidebar.selectbox("Select Session", session_options, on_change=reset_navigation)
    
    if not session:
        st.info("Please select a session.")
        return
    
    # Add cursor legend to sidebar
    st.sidebar.markdown("---")
    st.sidebar.markdown("### Legend")
    legend_img = create_cursor_legend()
    with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp_file:
        legend_path = tmp_file.name
        legend_img.save(legend_path)
    st.sidebar.image(legend_path, use_container_width=True)
    os.unlink(legend_path)  # Clean up temp file
    
    # Load data for selected session
    metadata, action_data, screenshots = load_session_data(category, website, session)
    
    # Display metadata
    with st.expander("Session Metadata", expanded=False):
        st.json(metadata)
    
    # Group actions by type
    actions = action_data.get("actions", [])
    grouped_actions = group_actions_by_type(actions)
    
    # Find before/after pairs
    action_pairs = find_before_after_pairs(actions)
    
    SKIP_ACTIONS = ["CURSOR_POSITIONING", "KEYBOARD_INPUT_MOVE_START"]

    # Filter out cursor positioning actions
    filtered_action_pairs = []
    for pair in action_pairs:
        before_action, after_action = pair
        
        # Skip pairs where either action type is in SKIP_ACTIONS
        should_skip = False
        if before_action and "type" in before_action:
            if before_action["type"] in SKIP_ACTIONS:
                should_skip = True
        if after_action and "type" in after_action:
            if after_action["type"] in SKIP_ACTIONS:
                should_skip = True
                
        if not should_skip:
            filtered_action_pairs.append(pair)
    
    # Replace original pairs with filtered ones
    action_pairs = filtered_action_pairs
    
    # Hardcode visualization settings to always be enabled
    show_gif = True
    show_cursor = True
    
    # Display session statistics at the top
    with st.expander("Session Statistics", expanded=False):
        st.markdown(f"**Total Actions:** {len(actions)}")
        st.markdown(f"**Total Screenshots:** {len(screenshots)}")
        st.markdown(f"**Action Pairs:** {len(action_pairs)}")
        st.markdown(f"**Session Duration:** {action_data.get('duration', 'N/A')} seconds")
        
        # Count of action types
        st.subheader("Action Types")
        action_types = {}
        for action in actions:
            action_type = action.get("type", "UNKNOWN")
            if action_type not in action_types:
                action_types[action_type] = 0
            action_types[action_type] += 1
        
        # Convert to list of tuples and sort by count
        action_type_counts = [(k, v) for k, v in action_types.items()]
        action_type_counts.sort(key=lambda x: x[1], reverse=True)
        
        for action_type, count in action_type_counts:
            st.text(f"{action_type}: {count}")
    
    # Slideshow controls
    st.subheader("Interaction Slideshow")
    
    # Create a session state for the current pair index if it doesn't exist
    if 'current_pair_index' not in st.session_state:
        st.session_state.current_pair_index = 0
    
    # Navigation controls
    col1, col2, col3 = st.columns([1, 3, 1])
    
    with col1:
        if st.button("Previous", key="prev_button"):
            st.session_state.current_pair_index = max(0, st.session_state.current_pair_index - 1)
    
    with col2:
        pair_slider = st.slider(
            "Navigate Actions", 
            min_value=0, 
            max_value=len(action_pairs) - 1 if action_pairs else 0, 
            value=st.session_state.current_pair_index,
            key="pair_slider"
        )
        st.session_state.current_pair_index = pair_slider
    
    with col3:
        if st.button("Next", key="next_button"):
            st.session_state.current_pair_index = min(len(action_pairs) - 1 if action_pairs else 0, st.session_state.current_pair_index + 1)
    
    # Display the current pair
    if action_pairs:
        current_pair = action_pairs[st.session_state.current_pair_index]
        before_action, after_action = current_pair
        
        # Display action details at the top
        with st.expander("Action Details", expanded=False):
            col1, col2 = st.columns(2)
            
            with col1:
                if before_action:
                    st.subheader("Before Action Details")
                    st.json(before_action)
            
            with col2:
                if after_action:
                    st.subheader("After Action Details")
                    st.json(after_action)
        
        # Display action descriptions
        col1, col2 = st.columns(2)
        
        # If showing GIF and we have both before and after images
        if show_gif and before_action and after_action and before_action.get("screenshot") and after_action.get("screenshot"):
            # Create a GIF and display it in a single column
            st.markdown(f"### Action: {after_action.get('description', 'N/A').replace('AFTER: ', '')}")
            
            before_screenshot = before_action.get("screenshot")
            after_screenshot = after_action.get("screenshot")
            
            if before_screenshot in screenshots and after_screenshot in screenshots:
                before_path = os.path.join(DATA_ROOT, category, website, session, session, "screenshots", before_screenshot)
                after_path = os.path.join(DATA_ROOT, category, website, session, session, "screenshots", after_screenshot)
                
                if os.path.exists(before_path) and os.path.exists(after_path):
                    # Create temporary directory for GIF
                    with tempfile.NamedTemporaryFile(suffix='.gif', delete=False) as tmp_file:
                        gif_path = tmp_file.name
                    
                    # Extract mouse positions if available
                    before_mouse_pos = None
                    after_mouse_pos = None
                    
                    if show_cursor:
                        if "mouse_position" in before_action and "screenshot" in before_action["mouse_position"]:
                            before_mouse_pos = before_action["mouse_position"]["screenshot"]
                        elif "mouse_position" in before_action and "screen" in before_action["mouse_position"]:
                            # Fallback to screen coords if screenshot coords not available
                            before_mouse_pos = before_action["mouse_position"]["screen"]
                        
                        if "mouse_position" in after_action and "screenshot" in after_action["mouse_position"]:
                            after_mouse_pos = after_action["mouse_position"]["screenshot"]
                        elif "mouse_position" in after_action and "screen" in after_action["mouse_position"]:
                            # Fallback to screen coords if screenshot coords not available
                            after_mouse_pos = after_action["mouse_position"]["screen"]
                    
                    # Add cursor to images if needed
                    if show_cursor:
                        before_img = Image.open(before_path)
                        after_img = Image.open(after_path)
                        
                        if before_mouse_pos:
                            before_img = draw_cursor(before_img, before_mouse_pos)
                            before_img.save(before_path + ".cursor.png")
                            before_path = before_path + ".cursor.png"
                        
                        if after_mouse_pos:
                            after_img = draw_cursor(after_img, after_mouse_pos)
                            after_img.save(after_path + ".cursor.png")
                            after_path = after_path + ".cursor.png"
                    
                    # Create and display the GIF
                    gif_path = create_transition_gif(before_path, after_path, gif_path)
                    if gif_path:
                        # Get image height for proper display
                        img_height = min(600, before_img.height if 'before_img' in locals() else 500)
                        
                        # Use components.html for better GIF rendering
                        with open(gif_path, "rb") as file:
                            gif_bytes = file.read()
                            gif_b64 = base64.b64encode(gif_bytes).decode("utf-8")
                            
                        html_code = f"""
                        <div style="display: flex; justify-content: center; margin-top: 20px;">
                            <img src="data:image/gif;base64,{gif_b64}" style="max-width: 100%; height: auto;">
                        </div>
                        """
                        
                        # Use the HTML component with specified height
                        components.html(html_code, height=img_height+50)
                        
                        # Clean up temporary cursor images if created
                        if show_cursor:
                            if os.path.exists(before_path) and before_path.endswith(".cursor.png"):
                                os.remove(before_path)
                            if os.path.exists(after_path) and after_path.endswith(".cursor.png"):
                                os.remove(after_path)
                    else:
                        # Fallback to separate images if GIF creation fails
                        st.warning("Could not create animation, showing separate images instead")
                        col1, col2 = st.columns(2)
                        with col1:
                            st.image(before_path, use_container_width=True)
                        with col2:
                            st.image(after_path, use_container_width=True)
                else:
                    st.warning(f"One or both screenshots not found: {before_path}, {after_path}")
            else:
                st.info("One or both screenshots are missing for this action pair.")
        else:
            # Show before and after separately
            with col1:
                if before_action:
                    st.markdown(f"### Before Action")
                    st.text(f"Action Type: {before_action.get('type', 'N/A')}")
                    
                    # Get the screenshot for the before action
                    before_screenshot = before_action.get("screenshot")
                    if before_screenshot and before_screenshot in screenshots:
                        screenshot_path = os.path.join(DATA_ROOT, category, website, session, session, "screenshots", before_screenshot)
                        if os.path.exists(screenshot_path):
                            if show_cursor and "mouse_position" in before_action:
                                # Prefer screenshot coordinates but fall back to screen coordinates
                                if "screenshot" in before_action["mouse_position"]:
                                    mouse_pos = before_action["mouse_position"]["screenshot"]
                                elif "screen" in before_action["mouse_position"]:
                                    mouse_pos = before_action["mouse_position"]["screen"]
                                img = Image.open(screenshot_path)
                                img = draw_cursor(img, mouse_pos)
                                
                                # Save to a temporary file
                                with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp_file:
                                    cursor_path = tmp_file.name
                                img.save(cursor_path)
                                
                                # Display the image with cursor
                                st.image(cursor_path, use_container_width=True)
                                
                                # Clean up
                                os.unlink(cursor_path)
                            else:
                                st.image(screenshot_path, use_container_width=True)
                        else:
                            st.warning(f"Screenshot not found: {screenshot_path}")
                    else:
                        st.info("No screenshot available for this action.")
            
            with col2:
                if after_action:
                    st.markdown(f"### Action: {after_action.get('description', 'N/A').replace('AFTER: ', '')}")
                    st.text(f"Action Type: {after_action.get('type', 'N/A')}")
                    
                    # Get the screenshot for the after action
                    after_screenshot = after_action.get("screenshot")
                    if after_screenshot and after_screenshot in screenshots:
                        screenshot_path = os.path.join(DATA_ROOT, category, website, session, session, "screenshots", after_screenshot)
                        if os.path.exists(screenshot_path):
                            if show_cursor and "mouse_position" in after_action:
                                # Prefer screenshot coordinates but fall back to screen coordinates
                                if "screenshot" in after_action["mouse_position"]:
                                    mouse_pos = after_action["mouse_position"]["screenshot"]
                                elif "screen" in after_action["mouse_position"]:
                                    mouse_pos = after_action["mouse_position"]["screen"]
                                img = Image.open(screenshot_path)
                                img = draw_cursor(img, mouse_pos)
                                
                                # Save to a temporary file
                                with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp_file:
                                    cursor_path = tmp_file.name
                                img.save(cursor_path)
                                
                                # Display the image with cursor
                                st.image(cursor_path, use_container_width=True)
                                
                                # Clean up
                                os.unlink(cursor_path)
                            else:
                                st.image(screenshot_path, use_container_width=True)
                        else:
                            st.warning(f"Screenshot not found: {screenshot_path}")
                    else:
                        st.info("No screenshot available for this action.")
                else:
                    st.info("No 'after' action available for this pair.")
    else:
        st.warning("No action pairs found for this session.")

if __name__ == "__main__":
    main()