data_loader: auto-download ground_truth from HF Dataset at startup

app/utils/data_loader.py

@@ -66,7 +66,24 @@ def ensure_ground_truth(root: str) -> str:
     
     # Try extracting from ground_truth.zip
     zip_path = os.path.join(root, "ground_truth.zip")
-    if os.path.exists(zip_path) and os.path.getsize(zip_path) > 10000:
+    
+    # If not local, try downloading from HF Dataset (for HF Space deployment)
+    if not (os.path.exists(zip_path) and os.path.getsize(zip_path) > 10000):
+        try:
+            from huggingface_hub import hf_hub_download
+            logger.info("Downloading ground_truth.zip from HF Dataset Valmbd/petimot-ground-truth ...")
+            zip_path = hf_hub_download(
+                repo_id="Valmbd/petimot-ground-truth",
+                filename="ground_truth.zip",
+                repo_type="dataset",
+                local_dir=root,
+            )
+            logger.info(f"Downloaded ground_truth.zip: {os.path.getsize(zip_path)//1e6:.0f} MB")
+        except Exception as e:
+            logger.warning(f"Could not download ground_truth from dataset: {e}")
+            _gt_extracted_flag[gt_dir] = True
+            return gt_dir
+
         logger.info(f"Extracting ground_truth.zip ({os.path.getsize(zip_path)//1e6:.0f} MB)...")
         os.makedirs(gt_dir, exist_ok=True)
         try: