"MinkowskiGenerativeConvolutionTranspose" function does not expand coordinates

[chrockey]

Hi,

I have tried to migrate my code to MinkowskiEngine V0.5.
Following the migration guide, I used ME.MinkowskiGenerativeConvolutionTranspose for a generative trans. conv. but it did not expand new coordinates.
However, ME.MinkowskiConvolutionTranspose with "expand_coordinates=True" did expand new coordinates successfully.
I attached my code.

  5 def test_union_map():                                                                                                                                                                                                                                                                 
  6     device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')                                                                                                                                                                                               
  7                                                                                                                                                                                                                                                                                       
  8     A = ME.SparseTensor(                                                                                                                                                                                                                                                              
  9             features=torch.FloatTensor([                                                                                                                                                                                                                                              
 10                 [1, 1],                                                                                                                                                                                                                                                               
 11                 [1, 1],                                                                                                                                                                                                                                                               
 12                 [1, 1],                                                                                                                                                                                                                                                               
 13                 [1, 1]                                                                                                                                                                                                                                                                
 14             ]),                                                                                                                                                                                                                                                                       
 15             coordinates=torch.IntTensor([                                                                                                                                                                                                                                             
 16                 [0, 0, 0, 1],                                                                                                                                                                                                                                                         
 17                 [0, 0, 1, 0],                                                                                                                                                                                                                                                         
 18                 [0, 0, 2, 0],                                                                                                                                                                                                                                                         
 19                 [0, 0, 3, 0]                                                                                                                                                                                                                                                          
 20             ]),                                                                                                                                                                                                                                                                       
 21             device=device                                                                                                                                                                                                                                                             
 22         )                                                                                                                                                                                                                                                                             
 23                                                                                                                                                                                                                                                                                       
 24     convtr = ME.MinkowskiGenerativeConvolutionTranspose(in_channels=2,                                                                                                                                                                                                                
 25                                                         out_channels=2,                                                                                                                                                                                                               
 26                                                         kernel_size=3,                                                                                                                                                                                                                
 27                                                         stride=1,                                                                                                                                                                                                                     
 28                                                         bias=False,                                                                                                                                                                                                                   
 29                                                         dimension=A.dimension).to(device)                                                                                                                                                                                             
 30                                                                                                                                                                                                                                                                                       
 31     B = convtr(A)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
 32                                                                                                                                                                                                                                                                                       
 33     print('\n', B.C)                                                                                                                                                                                                                                                                    

The output is

Using MinkowskiConvolutionTranspose with "expand_coordinates=True", the output is

By the way, thanks for your work!