Update app.py

app.py

@@ -184,30 +184,31 @@ def _to_html_table(S: np.ndarray, names: List[str]) -> str:
 
 @spaces.GPU(duration=_gpu_duration_gallery)
 def batch_similarity(files: List[str], model_name: str, pooling: str):
-    # files is a list of filepaths from gr.Files
     paths = files or []
     if len(paths) < 2:
-        return "Upload at least 2 images", None
-
+        return "<em>Upload at least 2 images</em>", None
     if not torch.cuda.is_available():
-        raise RuntimeError("CUDA not available. Ensure Space hardware is ZeroGPU.")
+        raise RuntimeError("CUDA not available. Ensure ZeroGPU is selected.")
 
     model_id = MODELS[model_name]
+    imgs = _open_images_from_paths(paths)
     embs = []
-    for img in _open_images_from_paths(paths):
+    for img in imgs:
         e, _, _ = _extract_core(img, model_id, pooling, want_overlay=False)
         embs.append(e)
 
     if len(embs) < 2:
-        return "Failed to read or embed images", None
+        return "<em>Failed to read or embed images</em>", None
 
     X = np.vstack(embs).astype(np.float32)
     Xn = X / np.clip(np.linalg.norm(X, axis=1, keepdims=True), 1e-8, None)
     S = Xn @ Xn.T
 
+    # save CSV and build HTML table
     csv_path = os.path.join(tempfile.gettempdir(), f"cosine_{uuid4().hex}.csv")
     np.savetxt(csv_path, S, delimiter=",", fmt="%.6f")
-    return f"Computed {len(embs)} embeddings", csv_path
+    html = _to_html_table(S, paths)
+    return html, csv_path
 
 
 # ---------------------------
@@ -235,7 +236,7 @@ with gr.Blocks() as app:
         gr.Markdown("Upload multiple images. We compute a cosine similarity matrix on GPU and return a CSV.")
         # Input as Files so you can multi-upload, plus a Gallery preview
         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)
+        gal = gr.Gallery(label="Images", columns=4, height=360, allow_preview=True)
         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")
@@ -255,7 +256,7 @@ with gr.Blocks() as app:
             return imgs
 
         files_in.change(_preview, inputs=files_in, outputs=gallery_preview)
-        go.click(batch_similarity, [files_in, model_dd2, pooling2], [status, csv])
+        go.click(batch_similarity, [gal, model_dd2, pooling2], [status, csv])