gradio version

README.md

@@ -1,6 +1,7 @@
 ---
 title: Zoo3D (VGGT + open-vocabulary 3D detection)
 sdk: gradio
+sdk_version: 5.17.1
 app_file: app.py
 pinned: false
 ---

mvp.py

@@ -146,7 +146,11 @@ def check_weights():
         print(f"Downloaded {cropformer_name}...")
     else:
         print(f"{cropformer_name} already exists...")
-check_weights()
+#
+# IMPORTANT (HF Spaces):
+# Do NOT download large weights at import time (startup). We'll download lazily
+# when running detection/reconstruction that actually needs them.
+#
 
 def extract_text_feature(descriptions, clip_model, target_path):
     text_tokens = tokenizer.tokenize(descriptions).to(device)
@@ -430,6 +434,11 @@ def reconstruct(
 
     print("Running run_model...")
     with torch.no_grad():
+        # Ensure CropFormer weights exist if downstream pipeline is enabled
+        try:
+            check_weights()
+        except Exception as e:
+            print(f"Warning: could not ensure Mask2Former weights at startup: {e}")
         vggt_model, metric3d_model, _ = _init_models()
         predictions = run_model(target_dir, vggt_model, metric3d_model=metric3d_model)
 
@@ -787,6 +796,12 @@ def detect_objects(text_labels, target_dir, conf_thres, *viz_args):
     # Require non-empty text labels
     if not text_labels or not isinstance(text_labels, str) or len([l.strip() for l in text_labels.split(";") if l.strip()]) == 0:
         return None, "Please enter at least one text label (separated by ';')."
+
+    # Ensure CropFormer weights exist (if detection pipeline uses them)
+    try:
+        check_weights()
+    except Exception as e:
+        print(f"Warning: could not ensure Mask2Former weights: {e}")
     
     # 1. Run reconstruction first if needed (checking if predictions exist)
     predictions_path = os.path.join(target_dir, "predictions.npz")
@@ -1017,7 +1032,32 @@ with gr.Blocks(
             input_video = gr.Video(label="Upload Video", interactive=True)
             input_images = gr.File(file_count="multiple", label="Upload Images", interactive=True)
 
-            image_gallery = gr.Gallery(
+            def _safe_gallery(**kwargs):
+                # Gradio API differs between versions; HF Spaces may run Gradio 6.x.
+                # Retry by removing unsupported kwargs.
+                while True:
+                    try:
+                        return gr.Gallery(**kwargs)
+                    except TypeError as e:
+                        msg = str(e)
+                        # Typical: "got an unexpected keyword argument 'show_download_button'"
+                        bad = None
+                        import re
+                        m = re.search(r"unexpected keyword argument '([^']+)'", msg)
+                        if m:
+                            bad = m.group(1)
+                        if bad and bad in kwargs:
+                            kwargs.pop(bad)
+                            continue
+                        # Fallback: drop known version-sensitive args
+                        for k in ["show_download_button", "preview", "object_fit", "columns", "height"]:
+                            if k in kwargs:
+                                kwargs.pop(k)
+                                break
+                        else:
+                            raise
+
+            image_gallery = _safe_gallery(
                 label="Preview",
                 columns=4,
                 height="300px",