add approximate method

app.py

@@ -118,7 +118,10 @@ def display_predictions(
 
     cropped_images: List[Image.Image] = []
     captions: List[str] = []
-    for closest_id, distance in zip(nn_ids, nn_distances):
+    progress_bar = st.progress(0)
+
+    for i, (closest_id, distance) in enumerate(zip(nn_ids, nn_distances)):
+        progress_bar.progress((i + 1) / len(nn_ids))
         closest_logo = logo_data[closest_id]
 
         cropped_image = get_cropped_image(
@@ -143,12 +146,26 @@ st.sidebar.title("Logo Nearest Neighbors Demo")
 st.sidebar.write(
     "Get first 100 nearest neighbors for a random annotated logo.\n\n"
     "CLIP model is used to generate embeddings, and nearest neighbors "
-    "are computed using a brute-force approach (no approximation)."
+    "are computed either using a brute-force approach or with ANN."
+)
+logo_id = st.sidebar.number_input("logo ID", step=1) or None
+approximate = (
+    st.sidebar.checkbox(
+        "ANN (HNSW)",
+        value=False,
+        help="Display approximate neighbors (instead of real "
+        "neighbors computed using brute-force approach",
+    )
+    or None
 )
 nn_data = load_nn_data(
-    "https://static.openfoodfacts.org/data/logos/exact_100_neighbours.json.gz"
+    f"https://static.openfoodfacts.org/data/logos/{'hnsw_50_closest_neighbours' if approximate else 'exact_100_neighbours'}.json.gz"
 )
 logo_data = load_logo_data(
     "https://static.openfoodfacts.org/data/logos/logo_annotations.jsonl.gz"
 )
-display_predictions(logo_data=logo_data, nn_data=nn_data)
+if approximate:
+    st.write("Using approximate nearest neighbors method")
+else:
+    st.write("Using exact (brute-force) nearest neighbors method")
+display_predictions(logo_data=logo_data, nn_data=nn_data, logo_id=logo_id)