app.py

import os
import gradio as gr
from gradio_imageslider import ImageSlider
from loadimg import load_img
import spaces
from transformers import AutoModelForImageSegmentation
import torch
from torchvision import transforms

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

# Выбор устройства
device = "cuda" if torch.cuda.is_available() else "cpu"

# Загрузка модели
birefnet = AutoModelForImageSegmentation.from_pretrained(
    "briaai/RMBG-2.0", trust_remote_code=True
)
birefnet.to(device)

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

output_folder = 'output_images'
if not os.path.exists(output_folder):
    os.makedirs(output_folder)

def fn(image):
    im = load_img(image, output_type="pil")
    im = im.convert("RGB")
    origin = im.copy()
    image = process(im)    
    image_path = os.path.join(output_folder, "no_bg_image.png")
    image.save(image_path)
    return (image, origin), image_path

@spaces.GPU
def process(image):
    image_size = image.size
    input_images = transform_image(image).unsqueeze(0).to(device)
    # Предсказание
    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):
    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 = ImageSlider(label="RMBG-2.0", type="pil")
slider2 = ImageSlider(label="RMBG-2.0", type="pil")
image = gr.Image(label="Upload an image")
image2 = gr.Image(label="Upload an image", type="filepath")
text = gr.Textbox(label="Paste an image URL")
png_file = gr.File(label="output png file")

chameleon = load_img("giraffe.jpg", output_type="pil")
url = "http://farm9.staticflickr.com/8488/8228323072_76eeddfea3_z.jpg"

tab1 = gr.Interface(
    fn, inputs=image, outputs=[slider1, gr.File(label="output png file")], examples=[chameleon], api_name="image"
)

tab2 = gr.Interface(fn, inputs=text, outputs=[slider2, gr.File(label="output png file")], examples=[url], api_name="text")

tab3 = gr.Interface(process_file, inputs=image2, outputs=png_file, examples=["giraffe.jpg"], api_name="png")

demo = gr.TabbedInterface(
    [tab1, tab2], ["input image", "input url"], title = (
    "RMBG-2.0 for background removal <br>"
    "<span style='font-size:16px; font-weight:300;'>"
    "Background removal model developed by "
    "<a href='https://bria.ai' target='_blank'>BRIA.AI</a>, trained on a carefully selected dataset,<br> "
    "and is available as an open-source model for non-commercial use.</span><br>"
    "<span style='font-size:16px; font-weight:500;'> For testing upload your image and wait.<br>"
    "<a href='https://go.bria.ai/3ZCBTLH' target='_blank'>Commercial use license</a> | "
    "<a href='https://huggingface.co/briaai/RMBG-2.0' target='_blank'>Model card</a> | "
    "<a href='https://blog.bria.ai/brias-new-state-of-the-art-remove-background-2.0-outperforms-the-competition' target='_blank'>Blog</a>"
    "</span><br>"
    "<span style='font-size:16px; font-weight:300;'>"
    "API Endpoint available on: "
    "<a href='https://platform.bria.ai/console/api/image-editing' target='_blank'>Bria.ai</a>, "
    "<a href='https://fal.ai/models/fal-ai/bria/background/remove' target='_blank'>fal.ai</a><br>"
    "ComfyUI node is available here: "
    "<a href='https://github.com/Bria-AI/ComfyUI-BRIA-API' target='_blank'>ComfyUI Node</a><br>"
    "Purchase commercial weights for commercial use: "
    "<a href='https://go.bria.ai/3D5EGp0' target='_blank'>here</a>"
    "</span>"
)
)

if __name__ == "__main__":
    demo.launch(show_error=True, server_name="0.0.0.0", server_port=7860, share=True)