How should I save the model of PyTorch if I want it loadable by OpenCV dnn module

Question

I train a simple classification model by PyTorch and load it by opencv3.3, but it throw exception and say

OpenCV Error: The function/feature is not implemented (Unsupported Lua type) in readObject, file /home/ramsus/Qt/3rdLibs/opencv/modules/dnn/src/torch/torch_importer.cpp, line 797 /home/ramsus/Qt/3rdLibs/opencv/modules/dnn/src/torch/torch_importer.cpp:797: error: (-213) Unsupported Lua type in function readObject

Model definition

class conv_block(nn.Module):
    def __init__(self, in_filter, out_filter, kernel):
        super(conv_block, self).__init__()

        self.conv1 = nn.Conv2d(in_filter, out_filter, kernel, 1, (kernel - 1)//2)
        self.batchnorm = nn.BatchNorm2d(out_filter)
        self.maxpool = nn.MaxPool2d(2, 2)

    def forward(self, x):
        x = self.conv1(x)
        x = self.batchnorm(x)
        x = F.relu(x)
        x = self.maxpool(x)

        return x

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()

        self.conv1 = conv_block(3, 6, 3)
        self.conv2 = conv_block(6, 16, 3)
        self.fc1 = nn.Linear(16 * 8 * 8, 120)
        self.bn1 = nn.BatchNorm1d(120)
        self.fc2 = nn.Linear(120, 84)
        self.bn2 = nn.BatchNorm1d(84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = x.view(x.size()[0], -1)
        x = F.relu(self.bn1(self.fc1(x)))
        x = F.relu(self.bn2(self.fc2(x)))
        x = self.fc3(x)
        return x

This model use Conv2d, ReLU, BatchNorm2d, MaxPool2d and Linear layer only, every layers are supported by opencv3.3

I save it by state_dict

torch.save(net.state_dict(), 'cifar10_model')

Load it by c++ as

std::string const model_file("/home/some_folder/cifar10_model");

std::cout<<"read net from torch"<<std::endl;
dnn::Net net = dnn::readNetFromTorch(model_file);

I guess I save the model with the wrong way, what is the proper way to save the model of PyTorch in order to load using OpenCV? Thanks

Edit :

I use another way to save the model, but it cannot be loaded either

torch.save(net, 'cifar10_model.net')

Is this a bug?Or I am doing something wrong?

Answer 1

I found the answer, opencv3.3 do not support PyTorch (https://github.com/pytorch/pytorch) but pytorch (https://github.com/hughperkins/pytorch), it is a big surprise, I never know there are another version of pytorch exist(looks like a dead project, long time haven't updated), I hope they could mention which pytorch they support on wiki.