Perform the annotation after the image is resized to avoid losing line size.

app.py

@@ -73,16 +73,14 @@ def process(input_image: Image.Image, prompt, a_prompt, n_prompt, max_faces: int
         raise gr.Error("Please provide an image")
     try:
         if image_file_live_opt == 'file':
-            input_image = input_image.convert('RGB')
-            empty = generate_annotation(
-                np.array(input_image), max_faces, min_confidence)
-            visualization = Image.fromarray(empty)  # Save to help debug.
-            visualization = pad_image(visualization).resize((512, 512))
+            # Resize before annotation so that we can keep our line-widths consistent with the training data.
+            input_image = pad_image(input_image.convert('RGB')).resize((512, 512))
+            empty = generate_annotation(np.array(input_image), max_faces, min_confidence)
+            visualization = Image.fromarray(empty)
         elif image_file_live_opt == 'webcam':
             base64_img = live_conditioning['image']
             image_data = base64.b64decode(base64_img.split(',')[1])
-            visualization = Image.open(BytesIO(image_data)).convert(
-                'RGB').resize((512, 512))
+            visualization = Image.open(BytesIO(image_data)).convert('RGB').resize((512, 512))
         if seed == -1:
             seed = random.randint(0, 2147483647)
         generator = torch.Generator(device).manual_seed(seed)