Method Initialize

lesson5-TrainingVGG/Classification.cpp:14–58

}

void Classifier::Initialize(int _num_classes, std::string _pretrained_path){
    std::vector<int> cfg_d = {64, 64, -1, 128, 128, -1, 256, 256, 256, -1, 512, 512, 512, -1, 512, 512, 512, -1};
    auto net_pretrained = VGG(cfg_d,1000,true);
    vgg = VGG(cfg_d,_num_classes,true);
    torch::load(net_pretrained, _pretrained_path);
    torch::OrderedDict<std::string, at::Tensor> pretrained_dict = net_pretrained->named_parameters();
    torch::OrderedDict<std::string, at::Tensor> model_dict = vgg->named_parameters();

    for (auto n = pretrained_dict.begin(); n != pretrained_dict.end(); n++)
    {
        if (strstr((*n).key().data(), "classifier")) {
            continue;
        }
        model_dict[(*n).key()] = (*n).value();
    }

    torch::autograd::GradMode::set_enabled(false);  // make parameters copying possible
    auto new_params = model_dict; // implement this
    auto params = vgg->named_parameters(true /*recurse*/);
    auto buffers = vgg->named_buffers(true /*recurse*/);
    for (auto& val : new_params) {
        auto name = val.key();
        auto* t = params.find(name);
        if (t != nullptr) {
            t->copy_(val.value());
        }
        else {
            t = buffers.find(name);
            if (t != nullptr) {
                t->copy_(val.value());
            }
        }
    }
    torch::autograd::GradMode::set_enabled(true);
    try
    {
        vgg->to(device);
    }
    catch (const std::exception&e)
    {
        std::cout << e.what() << std::endl;
    }

    return;
}

void Classifier::Train(int num_epochs, int batch_size, float learning_rate, std::string train_val_dir, std::string image_type, std::string save_path){
    std::string path_train = train_val_dir+ "\\train";