Upload 2 files

app/model_utils.py

@@ -86,8 +86,9 @@ def load_seg_model():
 # ── ZoeDepth loader (CPU, HF Hub) ─────────────────────────────────────────────
 def load_zoe_model():
     """
-    Loads ZoeDepth (ZoeD_N) from HuggingFace Hub on CPU.
-    Uses the transformers pipeline for simplicity and reliability.
+    Loads ZoeDepth (ZoeD_N) from Hugging Face Hub on ZeroGPU.
+    Initialisation during application startup lets ZeroGPU place the pipeline
+    on CUDA before the decorated Gradio callback requests a real GPU.
     Model: Intel/zoedepth-nyu  (indoor/outdoor depth estimation)
     """
     global _zoe_model
@@ -96,16 +97,21 @@ def load_zoe_model():
 
     from transformers import pipeline as hf_pipeline
 
-    print("=== LOADING ZoeDepth MODEL ===", flush=True)
+    print("=== LOADING ZoeDepth MODEL ON CUDA ===", flush=True)
     _zoe_model = hf_pipeline(
         task="depth-estimation",
         model="Intel/zoedepth-nyu",
-        device=-1,
+        device=0,
     )
     print("=== ZoeDepth MODEL LOADED ===", flush=True)
     return _zoe_model
 
 
+# ZeroGPU emulates CUDA during application startup. Initialising here caches the
+# downloaded model and avoids a CPU-only model load on the first user request.
+load_zoe_model()
+
+
 # ── Preprocessing ──────────────────────────────────────────────────────────────
 def preprocess_image(image: Image.Image) -> np.ndarray:
     """PIL Image → normalised (1, 512, 512, 3) float32 array."""