深度学习模型部署：VS2017(c++) + libtorch(1.5.0) + opencv(4.0.1)

好不容易搞通的，记录一下以免忘记

1.准备工作

1.1 libtorch配置

类似配置opencv，添加包含目录和库目录

1.2 pt文件准备

本次测试的是基于vggnet的猫狗识别。

1.2.1 pth_to_pt

import torch
import torchvision.models as models
import os
import torch.nn as nn

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# 加载模型
path_state_dict = os.path.join("..","data", "VGG_16_weight.pth")
model = models.vgg16()
num_ftrs = model.classifier._modules["6"].in_features
model.classifier._modules["6"] = nn.Linear(num_ftrs, 2)
model.to(device)
pretrained_state_dict