genmake_bg_remove.py

# genmake_bg_remove.py
import cv2
import numpy as np
from rembg import remove  # BiRefNet for background removal
import torch
from ultralytics import YOLO  # For SAM2 (Ultralytics version)
from PIL import Image
import os


# 1. BiRefNet Background Removal (default option)
def remove_bg_birefnet(input_image_path):
    input_image = Image.open(input_image_path)
    output_image = remove(input_image)  # Using BiRefNet for background removal
    output_image_path = input_image_path.replace(".jpg", "_no_bg.png").replace(".png", "_no_bg.png")
    output_image.save(output_image_path)
    return output_image_path


# 2. SAM2 Background Removal (pro option)
def remove_bg_sam2(input_image_path):
    # Load SAM2 model (Ultralytics)
    model = YOLO("sam2.pt")  # Load pre-trained SAM2 model

    # Load input image
    image = cv2.imread(input_image_path)
    results = model(image)  # Perform segmentation

    # Post-process results (segment)
    segmentation_mask = results[0].masks  # Get segmentation mask (person segmentation)
    mask = segmentation_mask.numpy()

    # Create a masked image
    masked_image = cv2.bitwise_and(image, image, mask=mask.astype(np.uint8))
    
    # Save the output image
    output_image_path = input_image_path.replace(".jpg", "_masked.png").replace(".png", "_masked.png")
    cv2.imwrite(output_image_path, masked_image)
    return output_image_path


# 3. Hybrid Approach (BiRefNet first, then SAM2 for refinement)
def remove_bg_hybrid(input_image_path):
    temp_image_path = remove_bg_birefnet(input_image_path)
    return remove_bg_sam2(temp_image_path)


# Entry function to call based on the user's selection
def process_background(input_image_path, method="auto"):
    if method == "auto":
        return remove_bg_birefnet(input_image_path)  # BiRefNet default
    elif method == "pro":
        return remove_bg_sam2(input_image_path)  # SAM2 pro
    elif method == "hybrid":
        return remove_bg_hybrid(input_image_path)  # Hybrid method
    else:
        raise ValueError("Invalid method: Choose from 'auto', 'pro', 'hybrid'")


# Example usage:
# img_path = 'path_to_your_image.jpg'
# result_path = process_background(img_path, 'auto')  # 'auto', 'pro', or 'hybrid'
# print(f"Processed image saved at: {result_path}")