AttributeError: 'torch.Size' object has no attribute 'numel'

[Hansoluo]

When I use it to test my net, it reports an error:
File "/torchstat/compute_memory.py", line 57, in compute_Conv2d_memory mread = batch_size * (inp.size()[1:].numel() + num_params(module)) AttributeError: 'torch.Size' object has no attribute 'numel'

And I found in pytorch documents torch.numel , only tensor have the attribute 'numel'. So, I have no idea How can I get this to work.If anyone can help, thanks.

I use python 3.5 Pytorch 0.4.1.

[yaotc]

the same error .

[lyakaap]

I fixed this problem. See #8

[woaichipinngguo]

I encountered the error again. @Hansoluo @Swall0w

[Hansoluo]

@woaichipinngguo @yaotc I used this file compute_memory.py to replace original compute_memory.py. And it worked for me. @lyakaap, Thanks for your help.

[woaichipinngguo]

ConvTranspose2d and Upsample are not supported. @Hansoluo

[yaotc]

compute_memory.py

@Hansoluo @lyakaap Thank you all ~ It worked!

[Hansoluo]

@woaichipinngguo "Upsample" layer have no paraments when inference, so inference memory and Flops is zero. "ConvTranspose2d" maybe you can use this math equation to calculate by hand. Or you can use it as same as "Conv2d", just add this layer in script.

def compute_memory(module, inp, out):
    ...
    elif isinstance(module, nn.ConvTranspose2d):
         return compute_Conv2d_memory(module, inp, out)
    ....

[xiaoketongxue]

I used this file compute_memory.py to replace original compute_memory.py.
but
File "D:\Anaconda3\lib\site-packages\torchstat\compute_memory.py", line 57, in compute_Conv2d_memory
mwrite = out.numel()

AttributeError: 'torch.Size' object has no attribute 'numel'
This is why?