app.py

import gradio as gr
from loadimg import load_img
import spaces
from transformers import AutoModelForImageSegmentation
import torch
from torchvision import transforms
import os
import uuid
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):
    im = load_img(image, output_type="pil")
    im = im.convert("RGB")
    # Get the segmented image (RGBA)
    processed_image = process(im)
    
    # Generate a unique filename for the processed image
    unique_id = str(uuid.uuid4())[:8]
    output_path = f"output_{unique_id}.jpg"
    
    # Create a white background and properly composite with the RGBA image
    white_bg = Image.new("RGB", processed_image.size, (255, 255, 255))
    if processed_image.mode == 'RGBA':
        # Use the alpha channel as a mask for compositing
        white_bg.paste(processed_image, mask=processed_image.split()[3])  # The 4th channel is alpha
        white_bg.save(output_path, format="JPEG")
        # Return the composited image for display to match what's being downloaded
        return white_bg, output_path
    else:
        rgb_image = processed_image.convert("RGB")
        rgb_image.save(output_path, format="JPEG")
        return rgb_image, output_path

@spaces.GPU
def process(image):
    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)
    
    # Since we're outputting RGB instead of RGBA, create a composite
    # We'll keep the image with mask for display purposes
    result = image.copy()
    result.putalpha(mask)
    return result

def process_file(f):
    im = load_img(f, output_type="pil")
    im = im.convert("RGB")
    # Get the segmented image (RGBA)
    transparent = process(im)
    
    # Save as JPEG instead of PNG
    unique_id = str(uuid.uuid4())[:8]
    output_path = f"output_{unique_id}.jpg"
    
    # Create a white background and properly composite with the RGBA image
    white_bg = Image.new("RGB", transparent.size, (255, 255, 255))
    if transparent.mode == 'RGBA':
        # Use the alpha channel as a mask for compositing
        white_bg.paste(transparent, mask=transparent.split()[3])  # The 4th channel is alpha
        white_bg.save(output_path, format="JPEG")
    else:
        rgb_image = transparent.convert("RGB")
        rgb_image.save(output_path, format="JPEG")
    
    return output_path

# Using a single Blocks API instead of TabbedInterface to avoid compatibility issues
with gr.Blocks(title="Background Removal Tool") as demo:
    with gr.Tabs():
        with gr.Tab("Image Upload"):
            with gr.Row():
                image_upload = gr.Image(label="Upload an image")
            with gr.Row():
                submit_btn = gr.Button("Process Image")
            with gr.Row():
                output_image = gr.Image(label="Processed Image")
                output_file = gr.File(label="Download Processed Image")
            
            submit_btn.click(fn=fn, inputs=image_upload, outputs=[output_image, output_file])
        
        with gr.Tab("URL Input"):
            with gr.Row():
                url_input = gr.Textbox(label="Paste an image URL")
            with gr.Row():
                submit_url_btn = gr.Button("Process URL")
            with gr.Row():
                output_image_url = gr.Image(label="Processed Image")
                output_file_url = gr.File(label="Download Processed Image")
            
            submit_url_btn.click(fn=fn, inputs=url_input, outputs=[output_image_url, output_file_url])
        
        with gr.Tab("File Output"):
            with gr.Row():
                image_file_upload = gr.Image(label="Upload an image", type="filepath")
            with gr.Row():
                submit_file_btn = gr.Button("Process and Download")
            with gr.Row():
                output_file_path = gr.File(label="Download JPEG File")
            
            submit_file_btn.click(fn=process_file, inputs=image_file_upload, outputs=output_file_path)

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