How to normalize data with libtorch?

1

hello, everyone. i have a question about normalization with libtorch. In Pytorch help document, there shows
"
torch.nn.functional. normalize ( input , p=2 , dim=1 , eps=1e-12 , out=None )

but in libotrch, it shows there is no function in libtorch.

auto datases_input_normalize = torch::nn::functional::normalize(datasets_input, 2, 1,1e-12, None);

how to solve this problem? Or is there any function can be used to normalize in libtorch?

Thanks very much!

2

Which PyTorch version are you using?
This dummy code snippet works in the latest stable release:

cpp_source = """
    torch::Tensor my_norm(torch::Tensor input){
    
      auto out = torch::nn::functional::normalize(input);
      return out;
    }
    
"""

module = torch.utils.cpp_extension.load_inline(
    name="my_norm",
    cpp_sources=cpp_source,
    functions="my_norm",
    verbose=True,
)

x = torch.randn(10, 10)
out = module.my_norm(x)
3

Thanks for your reply, but i have problem with
“auto out = torch::nn::functional::normalize(input);”

it shows no normalize in torch::nn::functional.

Do you know why?

4

Have you solved this problem?

5 
auto mean = torch::tensor({ 0.485, 0.456, 0.406 }, torch::kFloat32).unsqueeze(1).unsqueeze(2).to(device);
auto std = torch::tensor({ 0.229, 0.224, 0.225 }, torch::kFloat32).unsqueeze(1).unsqueeze(2).to(device);
image_tensor = image_tensor.sub(mean).div(std);