File size: 10,947 Bytes
9c32a2c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 |
import warnings
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category=DeprecationWarning)
from PIL import Image
import matplotlib.pyplot as plt
import torch
from torchvision import transforms
from transformers import AutoModelForImageSegmentation
import numpy as np
from pathlib import Path
from tqdm import tqdm
import os
import gc
from .remove_background import RemoveBackground
from custom_logger import logger_config
class RemoveBackgroundAI(RemoveBackground):
def __init__(self, model_name='briaai/RMBG-2.0', device='cuda' if torch.cuda.is_available() else 'cpu', image_size=(1024, 1024)):
"""
Initialize the BackgroundRemover with a pre-trained model.
"""
super().__init__("remove_background_ai")
self.device = device
self.image_size = image_size
# Load the model
self.model = AutoModelForImageSegmentation.from_pretrained(model_name, trust_remote_code=True)
if device == 'cuda':
torch.set_float32_matmul_precision('high')
self.model.to(device)
self.model.eval()
# Define image transformations
self.transform = transforms.Compose([
transforms.Resize(image_size, antialias=True),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
def process(self, image_path, output_path=None, save_alpha=True):
"""
Remove background while preserving exact foreground position and size.
Args:
image_path (str or Path): Path to the input image
output_path (str or Path, optional): Path to save the output image
save_alpha (bool): If True, save with transparency (PNG RGBA), else black background
Returns:
PIL.Image: Processed image with background removed
"""
# Load and preprocess the image
image = Image.open(image_path).convert("RGB")
original_size = image.size
input_tensor = self.transform(image).unsqueeze(0).to(self.device)
# Generate the mask
with torch.no_grad():
preds = self.model(input_tensor)[-1].sigmoid().cpu()
pred = preds[0].squeeze()
mask_pil = transforms.ToPILImage()(pred)
mask = mask_pil.resize(image.size, Image.LANCZOS)
# Create result image preserving exact position and size
if save_alpha:
# Create RGBA image with transparency
result_image = Image.new("RGBA", original_size, (0, 0, 0, 0))
image_rgba = image.convert("RGBA")
# Apply mask to create transparency
mask_array = np.array(mask)
image_array = np.array(image_rgba)
# Set alpha channel based on mask
image_array[:, :, 3] = mask_array
result_image = Image.fromarray(image_array)
else:
# Create RGB image with black background
result_image = Image.new("RGB", original_size, (0, 0, 0))
# Paste original image using mask
result_image.paste(image, mask=mask)
# Save output if path provided
if output_path:
if isinstance(output_path, Path):
output_path = str(output_path)
if save_alpha:
# Ensure PNG extension for transparency
if not output_path.lower().endswith('.png'):
output_path = os.path.splitext(output_path)[0] + '.png'
result_image.save(output_path, format="PNG")
else:
result_image.save(output_path)
return result_image
def remove_background(self, image_path, output_path=None, crop=False, bg_color=(0, 0, 0), save_alpha=False, keep_size=True, skip_crop=False):
"""
Original method - kept for backward compatibility.
For preserving exact position/size, use remove_background_preserve_position() instead.
"""
# Load and preprocess the image
image = Image.open(image_path).convert("RGB")
original_size = image.size
input_tensor = self.transform(image).unsqueeze(0).to(self.device)
# Generate the mask
with torch.no_grad():
preds = self.model(input_tensor)[-1].sigmoid().cpu()
pred = preds[0].squeeze()
mask_pil = transforms.ToPILImage()(pred)
mask = mask_pil.resize(image.size)
# Create RGBA image with transparency
image_rgba = image.convert("RGBA")
temp_image = Image.new("RGBA", image.size, (0, 0, 0, 0))
temp_image.paste(image_rgba, mask=mask)
# Skip cropping if requested
if skip_crop:
print(f"Skipping cropping for {image_path}")
else:
# Crop if requested
if crop:
temp_image = self._crop_to_content(temp_image)
if keep_size:
# Paste cropped image back onto original canvas size
padded_image = Image.new("RGBA", original_size, (0, 0, 0, 0))
offset_x = (original_size[0] - temp_image.size[0]) // 2
offset_y = (original_size[1] - temp_image.size[1]) // 2
padded_image.paste(temp_image, (offset_x, offset_y))
temp_image = padded_image
# Save output
if output_path:
if isinstance(output_path, Path):
output_path = str(output_path)
if save_alpha:
if not output_path.lower().endswith('.png'):
output_path = os.path.splitext(output_path)[0] + '.png'
temp_image.save(output_path, format="PNG")
else:
bg_image = Image.new("RGB", temp_image.size, bg_color)
bg_image.paste(temp_image, mask=temp_image.split()[3])
bg_image.save(output_path)
return temp_image
def _crop_to_content(self, image):
"""
Crop the image to the bounding box of the non-transparent content.
"""
img_array = np.array(image)
alpha_channel = img_array[:, :, 3]
non_empty_columns = np.where(alpha_channel.max(axis=0) > 0)[0]
non_empty_rows = np.where(alpha_channel.max(axis=1) > 0)[0]
if len(non_empty_columns) > 0 and len(non_empty_rows) > 0:
crop_box = (
non_empty_columns.min(),
non_empty_rows.min(),
non_empty_columns.max() + 1,
non_empty_rows.max() + 1
)
return image.crop(crop_box)
return image
def cleanup(self):
"""
Clean up resources used by the model.
"""
if self.device == 'cuda':
self.model.to('cpu')
del self.model
if torch.cuda.is_available():
torch.cuda.empty_cache()
gc.collect()
print("Model unloaded and resources cleaned up")
def remove_background_batch(folder, output_path=None, save_alpha=True, preserve_position=True):
"""
Process all images in a folder and remove backgrounds while preserving position and size.
Args:
folder (str): Folder containing images to process
output_path (str): Output folder path
save_alpha (bool): If True, saves output as PNG with transparency
preserve_position (bool): If True, keeps foreground in exact original position
"""
remover = BackgroundRemover()
input_path = Path(folder)
# Create output directory if it doesn't exist
if output_path:
Path(output_path).mkdir(parents=True, exist_ok=True)
# Find all image files
image_files = []
for ext in ['jpg', 'jpeg', 'png', 'bmp', 'tiff', 'webp']:
image_files.extend(input_path.glob(f"*.{ext}"))
image_files.extend(input_path.glob(f"*.{ext.upper()}"))
print(f"Found {len(image_files)} images to process")
try:
for img_path in tqdm(image_files, desc="Removing Background", unit="image"):
try:
# Determine output filename
if output_path:
output_filename = os.path.basename(img_path)
if save_alpha and not output_filename.lower().endswith('.png'):
output_filename = os.path.splitext(output_filename)[0] + '.png'
output_file = os.path.join(output_path, output_filename)
else:
output_file = img_path
if preserve_position:
# Use the new method that preserves exact position
remover.remove_background_preserve_position(
image_path=img_path,
output_path=output_file,
save_alpha=save_alpha
)
else:
# Use original method with no cropping
remover.remove_background(
image_path=img_path,
output_path=output_file,
crop=False, # No cropping to preserve position
bg_color=(0, 0, 0),
save_alpha=save_alpha,
skip_crop=True
)
# print(f"✓ Processed: {os.path.basename(img_path)}")
except Exception as e:
print(f"✗ Error processing {img_path}: {str(e)}")
except KeyboardInterrupt:
print("\nProcessing interrupted by user")
finally:
remover.cleanup()
# Single image processing function
def process(image_path, output_path=None, save_alpha=True):
"""
Process a single image and remove background while preserving position and size.
Args:
image_path (str): Path to input image
output_path (str, optional): Path to save output image
save_alpha (bool): If True, saves with transparency
Returns:
PIL.Image: Processed image
"""
remover = RemoveBackgroundAI()
try:
result = remover.remove_background_preserve_position(
image_path=image_path,
output_path=output_path,
save_alpha=save_alpha
)
print(f"✓ Successfully processed: {os.path.basename(image_path)}")
return result
except Exception as e:
print(f"✗ Error processing {image_path}: {str(e)}")
return None
finally:
remover.cleanup()
# Example usage
if __name__ == "__main__":
# Process entire folder - preserves exact position and size
remove_background_batch(
folder="../CaptionCreator/media/puzzle_x_pic/",
output_path="../CaptionCreator/media/processed/",
save_alpha=True,
preserve_position=True
) |