Using ReLU without BatchNorm

[facebookresearch]

Using ReLU without BatchNorm

[btgraham]

I have added an standalone ReLU function.
Regards
Ben

[HeyangQin]

Using self.spatial_size= self.sparseModel.input_spatial_size(torch.LongTensor([1, 1])) with scn.ReLU() seems to cause torch.nn.modules.module.ModuleAttributeError: 'ReLU' object has no attribute 'input_spatial_size' Error.

[btgraham]

Update pushed, thanks.

[aradhyamathur]

I implemented a custom print layer to print the shape of the output. I also received the above error. The following is the implementation.

class PrintLayer(nn.Module):
    def __init__(self):
        super().__init__()
    
    def forward(self, x ):
        print(x.shape)
        return x


model = scn.Sequential().add(
    scn.SubmanifoldConvolution(3, 24, 32, 3, False)
).add(
    PrintLayer()
).add(
    scn.BatchNormReLU(32)
).to(device)

Upon initialising as shown in the example leads to

inputSpatialSize = model.input_spatial_size(torch.LongTensor([512, 512, 512]))


---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-58-2f931c1817d2> in <module>
----> 1 inputSpatialSize = model.input_spatial_size(torch.LongTensor([512, 512, 512]))

~/anaconda3/envs/sparse_conv/lib/python3.6/site-packages/sparseconvnet-0.2-py3.6-linux-x86_64.egg/sparseconvnet/sequential.py in input_spatial_size(self, out_size)
     11     def input_spatial_size(self, out_size):
     12         for m in reversed(self._modules):
---> 13             out_size = self._modules[m].input_spatial_size(out_size)
     14         return out_size
     15 

~/anaconda3/envs/sparse_conv/lib/python3.6/site-packages/torch/nn/modules/module.py in __getattr__(self, name)
    574                 return modules[name]
    575         raise AttributeError("'{}' object has no attribute '{}'".format(
--> 576             type(self).__name__, name))
    577 
    578     def __setattr__(self, name, value):

AttributeError: 'PrintLayer' object has no attribute 'input_spatial_size'

Is this error related to the previous problem pointed out ?