app.py

import numpy as np
import torch
from torchvision.transforms.functional import normalize
import torch.nn.functional as F
from skimage import io
from PIL import Image, ImageFilter
from transformers import AutoModelForImageSegmentation, AutoImageProcessor, AutoModelForDepthEstimation
import streamlit as st

# Load models
segmentation_model = AutoModelForImageSegmentation.from_pretrained("briaai/RMBG-1.4", trust_remote_code=True)
depth_model = AutoModelForDepthEstimation.from_pretrained("depth-anything/Depth-Anything-V2-Small-hf")
depth_processor = AutoImageProcessor.from_pretrained("depth-anything/Depth-Anything-V2-Small-hf")

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
segmentation_model.to(device)
depth_model.to(device)

# Preprocessing function for segmentation
def preprocess_image(im: np.ndarray, model_input_size: list) -> torch.Tensor:
    if len(im.shape) < 3:
        im = im[:, :, np.newaxis]
    im_tensor = torch.tensor(im, dtype=torch.float32).permute(2, 0, 1)
    im_tensor = F.interpolate(torch.unsqueeze(im_tensor, 0), size=model_input_size, mode='bilinear')
    image = torch.divide(im_tensor, 255.0)
    image = normalize(image, [0.5, 0.5, 0.5], [1.0, 1.0, 1.0])
    return image

# Postprocessing function for segmentation mask
def postprocess_image(result: torch.Tensor, im_size: list) -> np.ndarray:
    result = torch.squeeze(F.interpolate(result, size=im_size, mode='bilinear'), 0)
    ma = torch.max(result)
    mi = torch.min(result)
    result = (result - mi) / (ma - mi)
    im_array = (result * 255).permute(1, 2, 0).cpu().data.numpy().astype(np.uint8)
    im_array = np.squeeze(im_array)
    return im_array

# Streamlit UI
st.title("Blur Effects App")
st.markdown("Choose between Gaussian blur (segmentation-based) or depth-based lens blur.")

# File uploader
uploaded_file = st.file_uploader("Upload an image", type=["jpg", "png", "jpeg"])

if uploaded_file:
    # Load the uploaded image
    orig_image = Image.open(uploaded_file).convert("RGB")
    orig_im_size = orig_image.size
    orig_im_np = np.array(orig_image)

    # Display original image
    st.image(orig_image, caption="Uploaded Image", use_column_width=True)

    # Effect selection
    effect_option = st.radio("Choose an effect", ("Gaussian Blur (Segmentation)", "Lens Blur (Depth Map)"))

    if effect_option == "Gaussian Blur (Segmentation)":
        # Gaussian Blur (Segmentation-Based)
        st.subheader("Gaussian Blur (Segmentation)")
        # Preprocess image for segmentation
        model_input_size = [256, 256]  # Resize for model compatibility
        image = preprocess_image(orig_im_np, model_input_size).to(device)

        # Inference using the segmentation model
        with torch.no_grad():
            result = segmentation_model(image)

        # Postprocess result to generate mask
        result_image = postprocess_image(result[0][0], orig_im_size)
        pil_mask_im = Image.fromarray(result_image)

        # Create binary mask for background and foreground separation
        binary_mask = pil_mask_im.point(lambda p: 255 if p > 170 else 0, '1')

        # Gaussian blur for background
        blur_intensity = st.slider("Blur Intensity (σ)", min_value=1, max_value=30, value=15)
        blurred_background = orig_image.filter(ImageFilter.GaussianBlur(blur_intensity))

        # Combine blurred background and sharp foreground
        binary_mask_rgba = binary_mask.convert("L").resize(orig_image.size)
        foreground = Image.composite(orig_image, blurred_background, binary_mask_rgba)

        # Display results side by side
        st.image([orig_image, foreground], caption=["Original Image", "Blurred Background"], use_column_width=True)

    elif effect_option == "Lens Blur (Depth Map)":
        # Lens Blur (Depth-Based)
        st.subheader("Lens Blur (Depth Map)")
        # Resize image for depth model processing
        inputs = depth_processor(images=orig_image, return_tensors="pt")

        # Generate depth map
        with torch.no_grad():
            outputs = depth_model(**inputs)
            depth = outputs.predicted_depth.squeeze().cpu().numpy()
        depth = np.array(Image.fromarray(depth).resize(orig_image.size, resample=Image.BICUBIC))

        # Resize depth map back to the original image size
        depth_min, depth_max = np.min(depth), np.max(depth)

        manual_depth_min = depth_min/10
        manual_depth_max = depth_max 

    
        normalized_depth = (depth - depth_min) / (depth_max - depth_min)
        adjusted_depth = np.clip((normalized_depth * (manual_depth_max - manual_depth_min)) + manual_depth_min, 0, 1)

        # Create depth-based blur effect
        max_blur = st.slider("Max Blur Intensity", min_value=1, max_value=10, value=3)
        depth_array = ((1 - adjusted_depth) * max_blur).astype(np.uint8)

        # Create blurred images for depth-based blurring
        blurred_images = [orig_image.filter(ImageFilter.GaussianBlur(i)) for i in range(max_blur + 1)]
        final_image = Image.new("RGB", orig_image.size)

        # Apply depth-based blur pixel by pixel
        for y in range(orig_image.height):
            for x in range(orig_image.width):
                blur_level = min(max_blur, max(0, depth_array[y, x]))
                final_image.putpixel((x, y), blurred_images[blur_level].getpixel((x, y)))

        # Display results side by side
        st.image([orig_image, final_image], caption=["Original Image", "Lens Blur Image"], use_column_width=True)