Redesigned flow

app.py

@@ -183,7 +183,7 @@ hypar["model"] = ISNetDIS()
  # Build Model
 net = build_model(hypar, device)
 
-
+'''
 def inference(image):
     image_path = image
     
@@ -207,6 +207,40 @@ def inference(image):
     im_dark.putalpha(pil_processed_mask)
     
     return [cropped_signature_image, processed_mask, im_dark]
+'''
+
+def inference(image):
+    image_path = image
+    
+    # Load and predict the mask for the original image
+    image_tensor, orig_size = load_image(image_path, hypar) 
+    original_mask = predict(net, image_tensor, orig_size, hypar, device)
+    
+    # Crop the signature from the original image
+    cropped_signature_image = crop_signature(image_path, original_mask, 64)
+    
+    # Resize the cropped image to the expected input size for the model
+    resized_cropped_image = cropped_signature_image.resize(hypar["input_size"], Image.BILINEAR)
+    
+    # Convert the resized cropped image to a tensor for model input
+    resized_cropped_image_tensor, _ = load_image(resized_cropped_image, hypar)
+    
+    # Predict the mask for the resized cropped signature image
+    cropped_mask = predict(net, resized_cropped_image_tensor, orig_size, hypar, device)
+    
+    # Convert the refined cropped mask to a PIL image
+    pil_cropped_mask = Image.fromarray(cropped_mask).convert('L')
+    
+    # Apply the refined mask to the resized cropped signature image
+    im_rgba_cropped = resized_cropped_image.copy()
+    im_rgba_cropped.putalpha(pil_cropped_mask)
+    
+    # Create a dark image with the same size as the resized cropped signature image
+    im_dark_cropped = Image.new('RGB', im_rgba_cropped.size, (0, 0, 0))
+    im_dark_cropped.putalpha(pil_cropped_mask)
+    
+    # The outputs are now based on the resized cropped image
+    return [im_rgba_cropped, pil_cropped_mask, im_dark_cropped]
 
 
 title = "Mysign.id - Signature Background removal based on DIS"