It seems that the X-CUBE-AI code generated the wrong shape for the tensors. The correct shape of the input tensor for exemple should be (7, 6, 6, 1) but the generated code has (1, 6, 6, 7). This means that the strides are incorrect.how can i fix this?

[jean-michel.d]

Hello,

Is it possible to have more details? Is it a ONNX model? X-CUBE-AI version?

The input tensor is a 4d tensor, if we not consider the batch dimension. what is the type of the operator in relation with this input tensor?

br,

JM

[jkayo.1]

Hello, 

Thanks for your help!.

I generate the files by converting ". H5 file" (using keras, tensorflow) using CUBEMX, the shape of the tensors was correct in this step after the generation of files (network.c,...), I realize that my input shape which declared in python was (Batch, Depth, Height, width, Channel) :(1, 7,6,6,1)was inverted to ( 1,6,6,7,1). 

All tensors in the model were inverted. I tried to manually change the shape of the tensors in network.c but it causes errors in the model creation.