update

app.py

@@ -157,6 +157,15 @@ def save_video(frames, path="output.mp4", fps=10):
     subprocess.run(ffmpeg_cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
     return path
 
+class InteractiveRunner:
+    def __init__(self, algo):
+        self.algo = algo
+
+    @spaces.GPU()
+    def run(self, first_frame, action, first_pose, curr_frame, device):
+        return self.algo.interactive(first_frame, action, first_pose, curr_frame, device=device)
+
+
 @hydra.main(
     version_base=None,
     config_path="configurations",
@@ -172,15 +181,17 @@ def run(cfg: DictConfig):
     
     memory_frames.append(load_image_as_tensor(DEFAULT_IMAGE))
 
-    @spaces.GPU()
-    def run_interactive(first_frame, action, first_pose, curr_frame, device):
-        global algo
-        new_frame = algo.interactive(first_frame,
-                                        action,
-                                        first_pose, 
-                                        curr_frame, 
-                                        device=device)
-        return new_frame
+    runner = InteractiveRunner(algo)
+
+    # @spaces.GPU()
+    # def run_interactive(first_frame, action, first_pose, curr_frame, device):
+    #     global algo
+    #     new_frame = algo.interactive(first_frame,
+    #                                     action,
+    #                                     first_pose, 
+    #                                     curr_frame, 
+    #                                     device=device)
+    #     return new_frame
 
     def set_denoising_steps(denoising_steps, sampling_timesteps_state):
         algo.sampling_timesteps = denoising_steps
@@ -198,11 +209,20 @@ def run(cfg: DictConfig):
         
         for i in range(len(actions)):
             memory_curr_frame += 1
-            new_frame = run_interactive(memory_frames[0],
-                                          actions[i],
-                                          None, 
-                                          memory_curr_frame, 
-                                          device=device)
+
+            # new_frame = run_interactive(memory_frames[0],
+            #                               actions[i],
+            #                               None, 
+            #                               memory_curr_frame, 
+            #                               device=device)
+
+            new_frame = runner.run(
+                        memory_frames[0],
+                        actions[i],
+                        None,
+                        memory_curr_frame,
+                        device
+                    )
 
             memory_frames.append(new_frame)
 
@@ -230,11 +250,20 @@ def run(cfg: DictConfig):
         memory_curr_frame = 0
         input_history = ""
 
-        _ = run_interactive(memory_frames[0],
-                                    actions[0],
-                                    poses[0],
-                                    memory_curr_frame,
-                                    device=device)
+        # _ = run_interactive(memory_frames[0],
+        #                             actions[0],
+        #                             poses[0],
+        #                             memory_curr_frame,
+        #                             device=device)
+
+        new_frame = runner.run(
+                        memory_frames[0],
+                        actions[0],
+                        poses[0],
+                        memory_curr_frame,
+                        device
+                    )
+
         return input_history, DEFAULT_IMAGE
 
     def on_image_click(SELECTED_IMAGE):
@@ -243,11 +272,13 @@ def run(cfg: DictConfig):
         reset()
         return SELECTED_IMAGE
 
-    _ = run_interactive(memory_frames[0],
-                                actions[0],
-                                poses[0],
-                                memory_curr_frame,
-                                device)
+    new_frame = runner.run(
+                        memory_frames[0],
+                        actions[0],
+                        poses[0],
+                        memory_curr_frame,
+                        device
+                    )
     
     print("first algo frame:", len(algo.frames))