环境配置

Visual Studio >= 2017 (2015试了很久一直报错）

CUDA >= 10.2

Pytorch 与 Libtorch 版本要一致

Qt的话要用MSVC2017编译器

模型转换

libtorch不依赖于python，python训练好的模型，需要转换为script model才能由libtorch加载，并进行推理。

Tracing 方法：

import torchimport torchvision# An instance of your model.model = torchvision.models.resnet18()# An example input you would normally provide to your model's forward() method.example = torch.rand(1, 3, 224, 224)# Use torch.jit.trace to generate a torch.jit.ScriptModule via tracing.traced_script_module = torch.jit.trace(model, example)traced_script_module.save("traced_resnet_model.pt")

下载Libtorch

，下载Libtorch,今天发现已经更新到1.9了都，前两天还是1.8.1.呢

注意使用的是Realsed版本还是Debug版本，下载完成后解压

VS中环境配置

类似于配置Opencv

将《your path to》\libtorch\include 和《your path to》\libtorch\include\torch\csrc\api\include 添加到C/C+±>附加包含目录

C/C+±>SDL检查->否 (/sdl-)

C/C+±>代码生成->运行库->多线程调试 (/MTd)

C/C+±>语言->符合模式->否

链接器-> 常规 ->附加库目录《your path to》\libtorch\lib

链接器->输入-> 附加依赖项

asmjit.lib;c10.lib;c10_cuda.lib;caffe2_detectron_ops_gpu.lib;caffe2_module_test_dynamic.lib;caffe2_nvrtc.lib;clog.lib;cpuinfo.lib;dnnl.lib;fbgemm.lib;libprotobufd.lib;libprotobuf-lited.lib;libprotocd.lib;mkldnn.lib;torch.lib;torch_cpu.lib;torch_cuda.lib;

Qt中环境配置

在pro文件末尾添加：

INCLUDEPATH += 《your path to》\opencv-4.5.0-vc14_vc15\opencv\build\include \《your path to》\libtorch17release\include \《your path to》\libtorch17release\include\torch\csrc\api\includeLIBS += -L《your path to》\opencv-4.5.0-vc14_vc15\opencv\build\x64\vc15\lib -lopencv_world450 \-L《your path to》\libtorch17release\lib -lc10 -ltorch -lc10_cuda -lcaffe2_detectron_ops_gpu -lc10d -ltorch_cpu \-ltorch_cuda -lgloo -lcaffe2_module_test_dynamic -lasmjit -lcaffe2_nvrtc -lclog -lcpuinfo -ldnnl -lfbgemm -lgloo_cuda \-lmkldnn -INCLUDE:?warp_size@cuda@at@@YAHXZ