app.py

import gradio as gr
from loadimg import load_img
import spaces
from transformers import AutoModelForImageSegmentation
import torch
from torchvision import transforms
from typing import Union, Tuple
from PIL import Image

torch.set_float32_matmul_precision(["high", "highest"][0])

birefnet = AutoModelForImageSegmentation.from_pretrained(
    "ZhengPeng7/BiRefNet", trust_remote_code=True
)
birefnet.to("cuda")

transform_image = transforms.Compose(
    [
        transforms.Resize((1024, 1024)),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
    ]
)

def fn(image: Union[Image.Image, str]) -> Tuple[Image.Image, Image.Image]:
    """
    Remove the background from an image and return both the transparent version and the original.

    This function performs background removal using a BiRefNet segmentation model. It is intended for use
    with image input (either uploaded or from a URL). The function returns a transparent PNG version of the image
    with the background removed, along with the original RGB version for comparison.

    Args:
        image (PIL.Image or str): The input image, either as a PIL object or a filepath/URL string.

    Returns:
        tuple:
            - processed_image (PIL.Image): The input image with the background removed and transparency applied.
            - origin (PIL.Image): The original RGB image, unchanged.
    """
    im = load_img(image, output_type="pil")
    im = im.convert("RGB")
    origin = im.copy()
    processed_image = process(im)
    return (processed_image, origin)

@spaces.GPU
def process(image: Image.Image) -> Image.Image:
    """
    Apply BiRefNet-based image segmentation to remove the background.

    This function preprocesses the input image, runs it through a BiRefNet segmentation model to obtain a mask,
    and applies the mask as an alpha (transparency) channel to the original image.

    Args:
        image (PIL.Image): The input RGB image.

    Returns:
        PIL.Image: The image with the background removed, using the segmentation mask as transparency.
    """
    image_size = image.size
    input_images = transform_image(image).unsqueeze(0).to("cuda")
    # Prediction
    with torch.no_grad():
        preds = birefnet(input_images)[-1].sigmoid().cpu()
    pred = preds[0].squeeze()
    pred_pil = transforms.ToPILImage()(pred)
    mask = pred_pil.resize(image_size)
    image.putalpha(mask)
    return image

def process_file(f: str) -> str:
    """
    Load an image file from disk, remove the background, and save the output as a transparent PNG.

    Args:
        f (str): Filepath of the image to process.

    Returns:
        str: Path to the saved PNG image with background removed.
    """
    name_path = f.rsplit(".", 1)[0] + ".png"
    im = load_img(f, output_type="pil")
    im = im.convert("RGB")
    transparent = process(im)
    transparent.save(name_path)
    return name_path

slider1 = gr.ImageSlider(label="Processed Image", type="pil", format="png")
slider2 = gr.ImageSlider(label="Processed Image from URL", type="pil", format="png")
image_upload = gr.Image(label="Upload an image")
image_file_upload = gr.Image(label="Upload an image", type="filepath")
url_input = gr.Textbox(label="Paste an image URL")
output_file = gr.File(label="Output PNG File")

# Example images
chameleon = load_img("butterfly.jpg", output_type="pil")
url_example = "https://hips.hearstapps.com/hmg-prod/images/gettyimages-1229892983-square.jpg"

tab1 = gr.Interface(fn, inputs=image_upload, outputs=slider1, examples=[chameleon], api_name="image")
tab2 = gr.Interface(fn, inputs=url_input, outputs=slider2, examples=[url_example], api_name="text")
tab3 = gr.Interface(process_file, inputs=image_file_upload, outputs=output_file, examples=["butterfly.jpg"], api_name="png")

demo = gr.TabbedInterface(
    [tab1, tab2, tab3], ["Image Upload", "URL Input", "File Output"], title="Background Removal Tool"
)

if __name__ == "__main__":
    demo.launch(show_error=True, mcp_server=True)