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("cpu")

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]:
    im = load_img(image, output_type="pil")
    im = im.convert("RGB")
    origin = im.copy()
    processed_image = process(im)
    return (origin, processed_image)

@spaces.GPU
def process(image: Image.Image) -> Image.Image:
    image_size = image.size
    input_images = transform_image(image).unsqueeze(0).to("cpu")
    # 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

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")
url_input = gr.Textbox(label="Paste an image URL")

tab1 = gr.Interface(fn, inputs=image_upload, outputs=slider1, api_name="image")
tab2 = gr.Interface(fn, inputs=url_input, outputs=slider2, api_name="text")

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

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