feat: implement Python BiRefNet background segmentation processor and installer

backend/process.py

@@ -0,0 +1,106 @@
+import sys
+import os
+import argparse
+
+# Force output buffering off so Electron gets real-time status prints
+sys.stdout.reconfigure(line_buffering=True)
+
+print("[STATUS] Importing libraries...")
+try:
+    import torch
+    from torchvision import transforms
+    from PIL import Image
+    from transformers import AutoModelForImageSegmentation
+except Exception as e:
+    print(f"[ERROR] Dependency import failed: {str(e)}", file=sys.stderr)
+    sys.exit(1)
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--input', required=True, help='Path to input image')
+    parser.add_argument('--output', required=True, help='Path to save output image')
+    parser.add_argument('--model', default='ZhengPeng7/BiRefNet', help='BiRefNet model name')
+    parser.add_argument('--device', default='auto', help='Device (cuda, cpu, auto)')
+    args = parser.parse_args()
+
+    if not os.path.exists(args.input):
+        print(f"[ERROR] Input file {args.input} does not exist", file=sys.stderr)
+        sys.exit(1)
+
+    print("[STATUS] Detecting device...")
+    if args.device == 'auto':
+        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    else:
+        device = args.device
+    print(f"[STATUS] Using device: {device.upper()}")
+
+    print("[STATUS] Loading BiRefNet model (this may take a moment on first run)...")
+    try:
+        # Load the model with trust_remote_code=True
+        model = AutoModelForImageSegmentation.from_pretrained(args.model, trust_remote_code=True)
+        model.to(device)
+        model.eval()
+    except Exception as e:
+        print(f"[ERROR] Failed to load model: {str(e)}", file=sys.stderr)
+        sys.exit(1)
+
+    print("[STATUS] Loading and preprocessing image...")
+    try:
+        orig_img = Image.open(args.input)
+        
+        # Keep original image color profile and transparency if any, but convert to RGB for model
+        if orig_img.mode != 'RGB':
+            img = orig_img.convert('RGB')
+        else:
+            img = orig_img
+
+        # BiRefNet works best with 1024x1024 input
+        transform_image = transforms.Compose([
+            transforms.Resize((1024, 1024)),
+            transforms.ToTensor(),
+            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+        ])
+        
+        input_tensor = transform_image(img).unsqueeze(0).to(device)
+    except Exception as e:
+        print(f"[ERROR] Failed to preprocess image: {str(e)}", file=sys.stderr)
+        sys.exit(1)
+
+    print("[STATUS] Running background removal inference...")
+    try:
+        with torch.no_grad():
+            preds = model(input_tensor)[-1].sigmoid().cpu()
+        
+        pred = preds[0].squeeze()
+    except Exception as e:
+        print(f"[ERROR] Inference failed: {str(e)}", file=sys.stderr)
+        sys.exit(1)
+
+    print("[STATUS] Generating transparency mask...")
+    try:
+        # Resize mask back to original image size
+        mask = transforms.ToPILImage()(pred).resize(orig_img.size, Image.Resampling.BILINEAR)
+
+        # Convert original image to RGBA and set the transparency alpha channel
+        rgba_img = orig_img.convert("RGBA")
+        rgba_img.putalpha(mask)
+    except Exception as e:
+        print(f"[ERROR] Mask generation failed: {str(e)}", file=sys.stderr)
+        sys.exit(1)
+
+    print(f"[STATUS] Saving output transparent image...")
+    try:
+        # Create output directory if it doesn't exist
+        out_dir = os.path.dirname(args.output)
+        if out_dir and not os.path.exists(out_dir):
+            os.makedirs(out_dir)
+            
+        rgba_img.save(args.output, "PNG")
+    except Exception as e:
+        print(f"[ERROR] Failed to save output image: {str(e)}", file=sys.stderr)
+        sys.exit(1)
+
+    print("[STATUS] DONE")
+
+if __name__ == '__main__':
+    main()