Update app.py

app.py

@@ -209,7 +209,6 @@ def batch_similarity(files: List[str], model_name: str, pooling: str):
 
 # ---------------------------
 # Image Classification using DINOv3 embeddings
-# Few shot nearest centroid on GPU
 # ---------------------------
 # State format:
 # state = {
@@ -304,12 +303,10 @@ def predict_class(image: Image.Image, model_name: str, pooling: str, state: Dict
 
     order = np.argsort(-probs)[: max(1, min(top_k, len(names)))]
     result_dict = {names[i]: float(probs[i]) for i in order}
-    # Full table for display
     full_table = "<ol>" + "".join(
         f"<li>{names[i]}  score {float(probs[i]):.4f}</li>" for i in order
     ) + "</ol>"
 
-    # gr.Label expects a dict of class to score for visualization
     return {"top_k": top_k, "prediction": names[order[0]]}, result_dict, full_table
 
 def clear_classes(_state: Dict):
@@ -339,15 +336,15 @@ with gr.Blocks() as app:
         }
         ```  """)
 
-    # ------------- Single -------------
+    # ------------- Embeddings -------------
     with gr.Tab("Embeddings"):
         with gr.Row():
             with gr.Column():
                 img = gr.Image(type="pil", label="Image", height=360)
                 model_dd = gr.Dropdown(choices=list(MODELS.keys()), value=DEFAULT_MODEL, label="Backbone")
                 pooling = gr.Radio(["CLS", "Mean of patch tokens"], value="CLS", label="Pooling")
-                overlay = gr.Checkbox(value=False, label="Show overlay")
-                run_btn = gr.Button("Extract on GPU")
+                overlay = gr.Checkbox(value=True, label="Show overlay")
+                run_btn = gr.Button("Extract")
             with gr.Column():
                 out_img = gr.Image(type="pil", label="Overlay or input", height=360)
                 preview = gr.Textbox(label="Embedding head", max_lines=2)
@@ -356,7 +353,6 @@ with gr.Blocks() as app:
 
         run_btn.click(extract_embedding, [img, model_dd, pooling, overlay], [out_img, preview, meta, download])
 
-        # Optional examples if you add files under ./examples
         ex_single = []
         for p in sorted(glob.glob("examples/*.*"))[:6]:
             ex_single.append([p, DEFAULT_MODEL, "CLS", False])
@@ -369,12 +365,12 @@ with gr.Blocks() as app:
 
     # ------------- Similarity -------------
     with gr.Tab("Similarity"):
-        gr.Markdown("Upload multiple images. We compute a cosine similarity matrix on GPU and return a CSV.")
+        gr.Markdown("Upload multiple images to compute a cosine similarity matrix and return a CSV.")
         files_in = gr.Files(label="Upload images", file_types=["image"], file_count="multiple", type="filepath")
         gallery_preview = gr.Gallery(label="Preview", columns=4, height=300)
         model_dd2 = gr.Dropdown(choices=list(MODELS.keys()), value=DEFAULT_MODEL, label="Backbone")
         pooling2 = gr.Radio(["CLS", "Mean of patch tokens"], value="CLS", label="Pooling")
-        go = gr.Button("Compute cosine on GPU")
+        go = gr.Button("Compute cosine")
         table = gr.HTML(label="Cosine similarity")
         csv = gr.File(label="cosine_similarity_matrix.csv")
 
@@ -403,13 +399,13 @@ with gr.Blocks() as app:
                 gr.Markdown("Build your labeled set by adding a few images per class.")
                 class_name = gr.Textbox(label="Class name")
                 class_files = gr.Files(label="Upload images for this class", file_types=["image"], type="filepath", file_count="multiple")
-                add_btn = gr.Button("Add class on GPU")
+                add_btn = gr.Button("Add class")
                 clear_btn = gr.Button("Clear classes")
                 state_view = gr.JSON(label="Classifier state")
             with gr.Column():
                 query_img = gr.Image(type="pil", label="Query image", height=360)
                 topk = gr.Slider(1, 10, value=3, step=1, label="Top K")
-                predict_btn = gr.Button("Predict on GPU")
+                predict_btn = gr.Button("Predict")
                 predicted = gr.Label(num_top_classes=3, label="Prediction")
                 scores_html = gr.HTML(label="Scores")