Allow 3D animation to run on GPU's other than GPU0

The device (potentially a CUDA device) wasn't being passed to the instantiation of the AdaBins InferenceHelper. Thus it was always picking the default -- typically "cuda:0".
This change passes the same device initialized in the main notebook (Which yeah, is hardcoded to "cuda:0".. but is really easy to change to something else).

disco_xform_utils.py

@@ -28,7 +28,7 @@ def transform_image_3d(img_filepath, midas_model, midas_transform, device, rot_m
         predictions using nyu dataset
         """
         print("Running AdaBins depth estimation implementation...")
-        infer_helper = InferenceHelper(dataset='nyu')
+        infer_helper = InferenceHelper(dataset='nyu', device=device)
 
         image_pil_area = w*h
         if image_pil_area > MAX_ADABINS_AREA:
@@ -128,4 +128,4 @@ def get_spherical_projection(H, W, center, magnitude,device):
     d = center - grid
     d_sum = torch.sqrt((d**2).sum(axis=-1))
     grid += d * d_sum.unsqueeze(-1) * magnitude 
-    return grid.unsqueeze(0)
+    return grid.unsqueeze(0)