dnn opencv c++ 'getMemoryShapes' 断言失败

Question

I have trained a CNN in keras saving the result as.h5 and.json.

After that I converted the files to a.pb using a python script, and now I want to import the network into opencv in c++ using cv::dnn::readNetFromTensorflow.

But I am getting an assertion failed on 'getMemoryShapes', what can be the issue here?

The error message is the following:

terminate called after throwing an instance of 'cv::Exception'
  what():  OpenCV(4.3.0) /build/opencv/src/opencv-4.3.0/modules/dnn/src/layers/convolution_layer.cpp:340: error: (-2:Unspecified error) Number of input channels should be multiple of 3 but got 32 in function 'getMemoryShapes'
Aborted (core dumped)

My code for now is very simple, I am just feeding the net with a sample image 32x32 since it is the input size of the network:

int main(int argc, char** argv) {

    cv::dnn::Net net;

    net = cv::dnn::readNetFromTensorflow("model.pb");

    cv::Mat image(32, 32, CV_8UC3, cv::Scalar(22,22,22));

    net.setInput(image);

    cv::Mat out;

    out = net.forward();

    return 0;
}

My input layer in keras is:

model.add(tf.keras.layers.Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(32, 32, 3)))

Answer 1

It can be fixed by finding the blob and forwarding it to the network:

   blob = cv::dnn::blobFromImage(image, 1.0, cv::Size(32,32));
   net.setInput(blob);