Upload folder using huggingface_hub

app.py

@@ -32,20 +32,20 @@ DISCLAIMER = (
 # Predict function
 # ---------------------------------------------------------------------------
 
+
 def predict(image: np.ndarray):
     """Called by Gradio on button click or example selection."""
     if image is None:
         empty = np.zeros((64, 64, 3), dtype=np.uint8)
         return (
-            empty,                     # annotated image
-            *([empty, "–"] * 5),       # 5× (nail crop, score label)
-            "–",                       # total
+            *([empty, "–"] * 5),  # 5× (nail crop, score label)
+            "–",  # total
             "Please upload an image.",
         )
 
     result = backend.predict(image)
 
-    outputs = [result["annotated_image"]]
+    outputs = []
     for nail_img, score in zip(result["nails"], result["napsi_scores"]):
         outputs.append(nail_img)
         outputs.append(str(score) if score >= 0 else "–")
@@ -68,7 +68,7 @@ with gr.Blocks(
     gr.Markdown(NAPSI_DESC)
     gr.Markdown(DISCLAIMER)
 
-    # ── Top row: input + annotated hand at equal size ─────────────────────
+    # ── Top row: input only (annotated hand preview removed) ──────────────
     with gr.Row():
         with gr.Column(scale=1):
             image_input = gr.Image(
@@ -81,37 +81,34 @@ with gr.Blocks(
             gr.Examples(
                 examples=[
                     ["assets/example_1.jpg"],
-                    ["assets/example_2.jpg"],
-                    ["assets/example_3.jpg"],
                 ],
                 inputs=image_input,
                 label="Example images",
             )
 
-        with gr.Column(scale=1):
-            annotated_out = gr.Image(
-                label="Detected hand",
-                type="numpy",
-                interactive=False,
-            )
-
     # ── Bottom rows: nail crops + scores ──────────────────────────────────
     with gr.Row():
-        nail_images = [gr.Image(label=f, type="numpy", interactive=False, height=160) for f in FINGER_NAMES]
+        nail_images = [
+            gr.Image(label=f, type="numpy", interactive=False, height=160)
+            for f in FINGER_NAMES
+        ]
     with gr.Row():
-        nail_scores = [gr.Textbox(label=f"NAPSI {f}", interactive=False) for f in FINGER_NAMES]
+        nail_scores = [
+            gr.Textbox(label=f"NAPSI {f}", interactive=False) for f in FINGER_NAMES
+        ]
 
     with gr.Row():
-        total_score = gr.Textbox(label="DeepNAPSI Total (one hand, 0–20)", interactive=False)
+        total_score = gr.Textbox(
+            label="DeepNAPSI Total (one hand, 0–20)", interactive=False
+        )
 
     error_box = gr.Textbox(label="Status", interactive=False, visible=True)
 
     # Wire outputs into a flat list matching predict() return order
-    all_outputs = (
-        [annotated_out]
-        + [x for pair in zip(nail_images, nail_scores) for x in pair]
-        + [total_score, error_box]
-    )
+    all_outputs = [x for pair in zip(nail_images, nail_scores) for x in pair] + [
+        total_score,
+        error_box,
+    ]
 
     predict_btn.click(fn=predict, inputs=image_input, outputs=all_outputs)
 
@@ -125,4 +122,3 @@ if __name__ == "__main__":
         favicon_path="assets/favicon-32x32.png",
         theme=gr.themes.Soft(),
     )
-

backend.py

@@ -8,6 +8,7 @@ Git LFS). If that file is absent it is downloaded from the HF Hub.
 
 from __future__ import annotations
 
+import concurrent.futures
 import os
 from pathlib import Path
 from typing import List
@@ -59,7 +60,12 @@ def _get_model_path() -> Path:
             "DeepNAPSIModel secret is set. Set one in the Space settings."
         )
     print(f"[backend] Downloading model from private repo {HF_REPO_ID} …")
-    path = hf_hub_download(HF_REPO_ID, HF_FILENAME, token=token)
+    # Run inside a ThreadPoolExecutor so that huggingface_hub's internal asyncio
+    # event loop is isolated; avoids the harmless-but-noisy
+    # "Invalid file descriptor: -1" Python 3.12 GC warning on Space startup.
+    with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
+        future = pool.submit(hf_hub_download, HF_REPO_ID, HF_FILENAME, token=token)
+        path = future.result()
     return Path(path)