update

algorithms/worldmem/df_video.py

@@ -807,7 +807,7 @@ class WorldMemMinecraft(DiffusionForcingBase):
             new_c2w_mat = euler_to_camera_to_world_matrix(first_pose)
             self_memory_c2w = new_c2w_mat[None, None].to(device)
             self_frame_idx = torch.tensor([[context_frames_idx]]).to(device)
-            return first_frame.cpu(), self_frames.cpu().numpy(), self_actions.cpu().numpy(), self_poses.cpu().numpy(), self_memory_c2w.cpu().numpy(), self_frame_idx.cpu().numpy()
+            return first_frame.cpu().numpy(), self_frames.cpu().numpy(), self_actions.cpu().numpy(), self_poses.cpu().numpy(), self_memory_c2w.cpu().numpy(), self_frame_idx.cpu().numpy()
         else:
             self_frames = torch.from_numpy(self_frames)
             self_actions = torch.from_numpy(self_actions).to(device)

app.py

@@ -241,7 +241,7 @@ def generate(keys):
 
         memory_frames.append(new_frame)
 
-    out_video = torch.stack(memory_frames)
+    out_video = np.stack(memory_frames)
     out_video = out_video.permute(0,2,3,1).numpy()
     out_video = np.clip(out_video, a_min=0.0, a_max=1.0)
     out_video = (out_video * 255).astype(np.uint8)