Update app.py

app.py

@@ -37,6 +37,7 @@ from threading import Thread
 from transformers import AutoModelForCausalLM
 from PIL import ImageDraw
 from torchvision.transforms.v2 import Resize
+from gradio_image_prompter import ImagePrompter
 
 os.environ["HF_TOKEN"] = os.environ.get("TOKEN_FROM_SECRET") or True
 moondream = AutoModelForCausalLM.from_pretrained(
@@ -204,6 +205,12 @@ def point(img, object):
         visible=True, value=img
     )
 
+@spaces.GPU(duration=10)
+def localized_query(img_prompt, question):
+    img, points = img_prompt["image"], img_prompt["points"]
+    yield repr(points), {"text": "..."}
+
+
 js = ""
 
 css = """
@@ -310,6 +317,18 @@ with gr.Blocks(title="moondream vl (new)", css=css, js=js) as demo:
                     submit.click(point, [img, prompt], [thought, ann])
                     prompt.submit(point, [img, prompt], [thought, ann])
                     img.change(point, [img, prompt], [thought, ann])
+                elif mode == "Localized":
+                    with gr.Group():
+                        with gr.Row():
+                            prompt = gr.Textbox(
+                                label="Input",
+                                value="What does the text say?",
+                                scale=4,
+                            )
+                            submit = gr.Button("Submit")
+                        img = ImagePrompter(show_label=False)
+                    submit.click(localized_query, [img, prompt], [output, thought])
+                    prompt.submit(localized_query, [img, prompt], [output, thought])
                 else:
                     gr.Markdown("Coming soon!")