Assert error conv2d in dims != weight dims

[eupston]

hi there,

I'm trying to run your code with my own model. The input shape for my prediction in python's keras is (1, 96, 114, 1), which works fine.

However when I create the same shaped Tensor in c++ and try to run the prediction I get:

ASSERT [src/layers/conv2d.cc:8] 'in.dims_[2] == weights_.dims_[3]' failed

my c++ code is:

int main() {
    auto model = Model::load("example.model");
    Tensor in{1, 96, 114, 1};
    Tensor out = model(in);
    out.print();
    return 0;
}

Any ideas?

[SangaRavi]

Hello,

Were you able to fix this issue? if so can you share the sample code?

Thanks.

[abdinf]

i am facing the same problem , in my python code i use this shape : data=data.reshape((1,28,28,1)) prediction = model.predict(data) and it workds in C++ i write: auto model = Model::load("example.model"); Tensor in{1, 28, 28, 1}; Tensor out = model(in); and i get the same error ? pliz any suggestions

[eupston]

Na I didn't end up solving it sorry. Decided to use: https://github.com/Dobiasd/frugally-deep which did the job for me.

[haxsender]

Hello! Got the same problem when trying to use Conv2D with padding='same'. Solved by seting padding='valid' before training.