Fix ZeroGPU: switch to demo.launch() + demo.app, drop gr.mount_gradio_app

app.py

@@ -5,7 +5,7 @@ import numpy as np
 from PIL import Image, UnidentifiedImageError
 import torch
 import gradio as gr
-from fastapi import FastAPI, File, UploadFile, HTTPException
+from fastapi import File, UploadFile, HTTPException
 from fastapi.responses import Response
 from transformers import DepthProImageProcessor, DepthProForDepthEstimation
 
@@ -42,8 +42,8 @@ def _run_depth_inference(image: Image.Image):
 
 
 # =====================================================================
-# 3. Gradio function - decorated with @spaces.GPU so ZeroGPU allocates
-#    a real GPU for this call.
+# 3. Gradio function - @spaces.GPU tells ZeroGPU to allocate a GPU here.
+#    This MUST be bound to a Gradio event handler for ZeroGPU to detect it.
 # =====================================================================
 @spaces.GPU(duration=120)
 def create_3d_point_cloud(input_pil_image):
@@ -104,17 +104,39 @@ def create_3d_point_cloud(input_pil_image):
 
 
 # =====================================================================
-# 4. FastAPI backend for the REST /get_metric_depth endpoint
+# 4. Gradio UI - uses demo.launch() (NOT gr.mount_gradio_app) so that
+#    ZeroGPU can properly detect the @spaces.GPU decorated function above.
 # =====================================================================
-fastapi_app = FastAPI(title="True Metric ToF Server")
+with gr.Blocks(title="Software-ToF 3D Cloud Visualizer") as demo:
+    gr.Markdown("# 📷 Software-ToF Camera Simulation")
+    gr.Markdown(
+        "Upload any image to calculate absolute metric depths and generate "
+        "an interactive 3D Point Cloud visualization."
+    )
+
+    with gr.Row():
+        with gr.Column():
+            input_img  = gr.Image(type="pil", label="Input Scene Photo")
+            submit_btn = gr.Button("Generate ToF 3D Structure", variant="primary")
+        with gr.Column():
+            output_3d = gr.Model3D(
+                label="Interactive 3D Point Cloud Canvas",
+                display_mode="point_cloud",
+            )
+
+    submit_btn.click(fn=create_3d_point_cloud, inputs=[input_img], outputs=[output_3d])
 
 
-@fastapi_app.get("/status")
+# =====================================================================
+# 5. Add the REST /get_metric_depth endpoint via Gradio's internal
+#    FastAPI app - this avoids gr.mount_gradio_app which breaks ZeroGPU.
+# =====================================================================
+@demo.app.get("/status")
 def status():
     return {"status": "running", "device": str(device)}
 
 
-@fastapi_app.post("/get_metric_depth")
+@demo.app.post("/get_metric_depth")
 async def get_metric_depth(file: UploadFile = File(...)):
     if not (file.content_type or "").startswith("image/"):
         raise HTTPException(status_code=400, detail="Uploaded file must be an image.")
@@ -140,31 +162,9 @@ async def get_metric_depth(file: UploadFile = File(...)):
 
 
 # =====================================================================
-# 5. Gradio UI
-# =====================================================================
-with gr.Blocks(title="Software-ToF 3D Cloud Visualizer") as demo:
-    gr.Markdown("# 📷 Software-ToF Camera Simulation")
-    gr.Markdown(
-        "Upload any image to calculate absolute metric depths and generate "
-        "an interactive 3D Point Cloud visualization."
-    )
-
-    with gr.Row():
-        with gr.Column():
-            input_img  = gr.Image(type="pil", label="Input Scene Photo")
-            submit_btn = gr.Button("Generate ToF 3D Structure", variant="primary")
-        with gr.Column():
-            output_3d = gr.Model3D(
-                label="Interactive 3D Point Cloud Canvas",
-                display_mode="point_cloud",
-            )
-
-    submit_btn.click(fn=create_3d_point_cloud, inputs=[input_img], outputs=[output_3d])
-
-
-# =====================================================================
-# 6. Mount Gradio onto FastAPI.
-#    Hugging Face (sdk: gradio) auto-discovers this `app` variable and
-#    serves it via uvicorn on port 7860 - no manual uvicorn.run() needed.
+# 6. Launch - Hugging Face auto-detects the `demo` variable when
+#    sdk: gradio is set and launches it. The explicit launch() call
+#    here ensures it also works when run locally.
 # =====================================================================
-app = gr.mount_gradio_app(fastapi_app, demo, path="/")
+if __name__ == "__main__":
+    demo.launch()