app.py

import gradio as gr
from transformers import pipeline
from PIL import Image, ImageFilter
import numpy as np
import io

# Load the segmentation pipeline
pipe = pipeline("image-segmentation", model="mattmdjaga/segformer_b2_clothes")

# Simplified refine_mask function
def refine_mask(mask):
    """Simplify and smooth the segmentation mask."""
    mask_array = np.array(mask)
    mask_array = (mask_array > 128).astype(np.uint8) * 255  # Threshold to binary mask
    refined_mask = Image.fromarray(mask_array).filter(ImageFilter.GaussianBlur(0.5))  # Smooth edges
    return refined_mask

# Function to blur the background
def blur_background(image: bytes, blur_radius: int) -> bytes:
    # Convert the image from bytes to PIL Image
    image = Image.open(io.BytesIO(image))

    # Perform segmentation
    result = pipe(image)
    
    # Extract the background mask
    background_mask = None
    for entry in result:
        if entry["label"] == "Background":
            background_mask = refine_mask(entry["mask"])  # Refine the background mask
            break
    
    if background_mask is None:
        return image  # If no background is detected, return the original image
    
    # Convert the image and mask to NumPy arrays
    image_np = np.array(image)
    background_mask_np = np.array(background_mask)
    
    # Create a blurred version of the entire image
    blurred_image = image.filter(ImageFilter.GaussianBlur(radius=blur_radius))
    blurred_np = np.array(blurred_image)
    
    # Combine the original image and the blurred background
    final_image = np.where(background_mask_np[..., None] == 255, blurred_np, image_np).astype(np.uint8)
    
    # Convert back to PIL image and then to bytes
    output_image = Image.fromarray(final_image)
    
    # Save the image to a bytes buffer
    img_byte_arr = io.BytesIO()
    output_image.save(img_byte_arr, format='PNG')
    img_byte_arr.seek(0)
    
    return img_byte_arr.read()

# Gradio Interface
def predict(image: bytes, blur_radius: int) -> bytes:
    return blur_background(image, blur_radius)

# API setup
def launch_api():
    # Expose the API
    interface = gr.Interface(
        fn=predict,
        inputs=[
            gr.Image(type="bytes"),  # Input image as bytes
            gr.Slider(1, 50, step=1, label="Blur Intensity")  # Slider for blur radius
        ],
        outputs=gr.Image(type="bytes"),  # Output image as bytes
        title="Background Blur API",
        description="This API blurs the background of an image while preserving the subject.",
    )

    interface.launch(share=True)  # Launch the Gradio interface as an API

# Launch the API
launch_api()