header only, deep learning framework with no dependencies other than libtorch

LinkOption(libtorch1.9.0):
-INCLUDE:?warp_size@cuda@at@@YAHXZ

LinkOption(libtorch1.12.1):
-INCLUDE:?warp_size@cuda@at@@YAHXZ -INCLUDE:?_torch_cuda_cu_linker_symbol_op_cuda@native@at@@YA?AVTensor@2@AEBV32@@Z

Enabled to build with libtorch1.5

Directory structure

This project aims to be a wrapper for libtorch to make tiny-dnn compatible with GPU. tiny-dnn really great, but unfortunately it can not be calculated on a GPU. At first glance, this header-only framework aims to be used as written in tiny-dnn.

Include path settings

Libtorch_path/include
Libtorch_path/include/torch/csrc/api/include
cpp_torch_path
cpp_torch_path/include

Library path setting

Libtorch_path/lib
cpp_torch_path

Minimum include file

#include "cpp_torch.h"

progress
tiny_dnn

tiny_dnn::progress_display disp(train_images.size());

cpp_torch

cpp_torch::progress_display disp(train_images.size());

cpp_torch::progress_display disp(train_images.size());

data set download
What you can do is still limited.

MNIS
CIFAR10
DCGAN

Do you have a favorite cute person?

This app was created with C # for GUI and C ++ only for core processing. Python is not required.

visual studio 2015,2017,2019

libtorch Please adapt the version of cuda to your environment

BSD 3-Clause License Copyright (c) 2013, Taiga Nomi

tiny_dnn was good, but unfortunately development has clearly stopped. Therefore, we created cpp_torch that can be used instead.

If you are building in C++,
Even if Python or pytorch (libtorch) is changed Should work.  

Will the Python app function correctly next month?  
Is there a guarantee that customers will not update python, etc.?