app.py

import streamlit as st
from streamlit_drawable_canvas import st_canvas
from PIL import Image
import numpy as np
import random
import vipainting
import time
import threading
from queue import Queue
import os

# Initialize queues to handle in-progress images for fitting and sampling
image_queue = Queue()
sampling_queue = Queue()

# Set up the app title
st.title("Mask Your Own Inpaint")

@st.cache_data
def load_images():
    """Loads images from an NPZ file only once to cache."""
    data = np.load("data/sflckr_all_images.npz")

    images = data["images"]
    return images

# Initialize session state to hold the selected random image index
if "random_idx" not in st.session_state:
    st.session_state.random_idx = None

# Load images and add button to randomly select one
images = load_images()
if st.button("Random Pick"):
    st.session_state.random_idx = random.randint(0, images.shape[0] - 1)

def run_inpainting(random_idx, mask_array, image_queue, sampling_queue):
    """Starts inpainting and sends images to the queue."""
    vipainting.vipaint(random_idx, mask_array, image_queue, sampling_queue)

# Only proceed if a random image has been selected
if st.session_state.random_idx is not None:
    img_array = images[st.session_state.random_idx]
    img_pil = Image.fromarray(img_array)

    # Set up drawing canvas and display columns
    col1, col2 = st.columns(2)
    with col1:
        st.write("Draw your mask on the image below:")
        canvas_result = st_canvas(
            fill_color="rgba(255, 0, 0, 0.3)",
            stroke_width=50,
            stroke_color="black",
            background_image=img_pil,
            update_streamlit=True,
            width=300,
            height=300,
            drawing_mode="freedraw",
            key="canvas"
        )

    # Process the mask if drawn
    if canvas_result.image_data is not None:
        mask = canvas_result.image_data[:, :, 3]
        binary_mask = (mask > 128).astype(np.uint8) * 255

        # Show the binary mask in the right column
        with col2:
            st.write("Masked Image")
            st.image(binary_mask, caption="Binary Mask", width=300)


        mask_image = Image.fromarray(binary_mask).resize((512, 512))  

        mask_array = 255 - np.array(mask_image)
        mask_array = np.expand_dims(mask_array, axis=(0, 1))

        

        # Inpaint button action
        if st.button("inpaint"):
            st.write("Please wait...")

            # Start inpainting in a separate thread
            inpaint_thread = threading.Thread(target=run_inpainting, args=(st.session_state.random_idx, mask_array, image_queue, sampling_queue))
            inpaint_thread.start()

            # Display initial image and segmentation map
            img_left, img_right = st.columns(2)
            img_left_placeholder = img_left.empty()
            img_right_placeholder = img_right.empty()
            img_left_placeholder.image(img_pil, caption="True Image", width=300)

            seg_image_path = f"results/{st.session_state.random_idx}/input.png"
            while not os.path.exists(seg_image_path):
                time.sleep(0.5)
            img_right_image = Image.open(seg_image_path)
            img_right_placeholder.image(img_right_image, caption="Segmentation Map", width=300)

            # Progress tracking for fitting
            expected_updates = 100
            progress_bar = st.progress(0)
            st.write("Fitting in progress")
            displayed_images = 0

            # Alternating display for fitting updates
            col_left, col_right = st.columns(2)
            left_placeholder = col_left.empty()
            right_placeholder = col_right.empty()

            while displayed_images < expected_updates:
                if not image_queue.empty():
                    img = image_queue.get()
                    if displayed_images % 2 == 0:
                        left_placeholder.image(img, caption=f"Progress Update {displayed_images + 1}", width=300)
                    else:
                        right_placeholder.image(img, caption=f"Progress Update {displayed_images + 1}", width=300)
                    displayed_images += 1
                    progress_bar.progress(displayed_images / expected_updates)
                time.sleep(0.3)

            # Progress tracking for sampling
            expected_updates = 10
            s_progress_bar = st.progress(0)
            displayed_images = 0
            st.write("Sampling in progress")
            sample_left, sample_right = st.columns(2)
            sleft_placeholder = sample_left.empty()
            sright_placeholder = sample_right.empty()

            while displayed_images < expected_updates:
                if not sampling_queue.empty():
                    img = sampling_queue.get()
                    if displayed_images % 2 == 0:
                        sleft_placeholder.image(img, caption=f"Sampling Update {displayed_images + 1}", width=300)
                    else:
                        sright_placeholder.image(img, caption=f"Sampling Update {displayed_images + 1}", width=300)
                    displayed_images += 1
                    s_progress_bar.progress(displayed_images / expected_updates)
                time.sleep(0.3)

            # Wait for inpainting to finish
            inpaint_thread.join()
            st.success("Inpainting completed!")