Use image url

app.py

@@ -6,14 +6,25 @@ import torch
 import typing
 import os
 import subprocess
+import requests
+import PIL
+
+
+def download_image(url) -> PIL.Image.Image:
+    """Download an image from a URL and return it as a PIL image."""
+    return PIL.Image.open(requests.get(url, stream=True).raw)
 
 
 def image_to_sam_image_embedding(
-    image: np.ndarray,
+    image_url: str,
     model_size: typing.Literal["base", "large", "huge"] = "base",
 ) -> str:
     """Generate an image embedding."""
 
+    image = download_image(image_url)
+    image = image.convert("RGB")
+    image = np.asarray(image)
+
     # Select model size
     if model_size == "base":
         predictor = base_predictor
@@ -77,10 +88,8 @@ if __name__ == "__main__":
     app = gr.Interface(
         fn=image_to_sam_image_embedding,
         inputs=[
-            gr.components.Image(label="Image URL"),
-            gr.components.Radio(
-                choices=["base", "large", "huge"], label="Model Size", default="base"
-            ),
+            gr.components.Textbox(label="Image URL"),
+            gr.components.Radio(choices=["base", "large", "huge"], label="Model Size"),
         ],
         outputs="text",
     )

requirements.txt

@@ -4,4 +4,5 @@ torchvision
 git+https://github.com/facebookresearch/segment-anything.git
 pycocotools
 onnxruntime
-onnx
+onnx
+Pillow