"Trying to copy blobs of different sizes" in caffe

[Michael Philippov]

Hello!
These are my dims in my deploy file.

layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param { shape: { dim: 64 dim: 1 dim: 28 dim: 28 } }
}

This is my code:

Введите код.    Net< float > net( "/home/phil/caffeData/lenet.prototxt", TEST );
    net.CopyTrainedLayersFrom("/home/phil/caffeData/_iter_5000.caffemodel");


    Datum datum;
    if( !ReadImageToDatum("/home/phil/caffeData/cat.jpg", 64, 28, 28, &datum) )
    {
        qDebug() << QObject::tr( "Невозможно прочесть изображение!" );
        return;
    }

    Blob< float > * blob = new Blob< float >(1, datum.channels(), datum.height(), datum.width());

    BlobProto proto;
    proto.set_num(1);
    proto.set_channels( datum.channels() );
    proto.set_height(datum.height());
    proto.set_width(datum.width());

    const int data_size = datum.channels() * datum.height() * datum.width();
    int size_in_datum = std::max<int>(datum.data().size(),
                                      datum.float_data_size());

    for (int i = 0; i < size_in_datum; ++i)
    {
        proto.add_data(0.);
    }

    const string& data = datum.data();
    if (data.size() != 0)
    {
        for (int i = 0; i < size_in_datum; ++i)
        {
            proto.set_data(i, proto.data(i) + (uint8_t)data[i]);
        }
    }

    blob->FromProto(proto);

    vector< Blob< float > * > bottom;
    bottom.push_back(blob);
    float type = 0.0;
    const vector< Blob< float > * >& result = net.Forward( bottom, &type );..

When I try to run it, I have a mistake like "

"Trying to copy blobs of different sizes". what's wrong?

[Przemek D]

Sounds like a typical error encountered when trying to load weights from a caffemodel generated for a different prototxt. Would be helpful if you attached the proto you attempt to load as well as the one you used to train your network.