Error loading mnist ONNX model

[lauracanalini]

Hi, we are trying to use onnx to save the eddl models, so we tried to run 2_onnx_save_net_to_file.cpp and load the file saved with 3_onnx_import_net_from_file.cpp. During loading we get this error:

Ir_version = 6
Operator domain  = 
Operator version = 11
Producer_name: EDDL
Producer_version: 0.1
Domain: 
Model_version: 0
Net imported from ONNX succesfully
Generating Random Table
Selecting GPU device 0
EDDLL is running on GPU device 0, GeForce RTX 2080 Ti
CuBlas initialized on GPU device 0, GeForce RTX 2080 Ti
CuRand initialized on GPU device 0, GeForce RTX 2080 Ti
terminate called after throwing an instance of 'std::runtime_error'
  what():  Incorrect padding type (PoolDescriptor::PoolDescriptor)
Aborted (core dumped)

System information:

• Ubuntu 19.04
• GCC 8.3.0
• eddl master branch (commit d89e085aed9bdb45d5b0b23fc9e46a87943795cb)

Does this also happen to you?

[chavicoski]

Can you share the code or the topology of the network?

[lauracanalini]

We're trying to use your examples, so the network is the one in eddl/examples/onnx/2_onnx_save_net_to_file.cpp.

// Define network
    layer in = Input({784});
    layer l = in;  // Aux var

     l = l=Reshape(l,{1,28,28});
     l=MaxPool(ReLu(BatchNormalization(Conv(l,32,{3,3},{1,1}))),{2,2});
     l=MaxPool(ReLu(BatchNormalization(Conv(l,64,{3,3},{1,1}))),{2,2});
     l=MaxPool(ReLu(BatchNormalization(Conv(l,128,{3,3},{1,1}))),{2,2});

     l=Flatten(l);

     l=Activation(Dense(l,256),"relu");

     l=Activation(Dense(l,128),"relu");

     layer out=Activation(Dense(l,num_classes),"softmax");

     // Net define input and output layers list
     model net=Model({in},{out});

[prittt]

Hey @chavicoski, is there any news from your side?

[chavicoski]

Not yet, we are working on it.

[prittt]

Ok, thank you. Let us know!

[RParedesPalacios]

Hi, Sorry for my previous examples. Effectively after git pull over master and re-build the new examples fails

[chavicoski]

The error is already fixed and uploaded to master. You can try to run it again