This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA4VM: Transposing data in TIDL

Part Number: TDA4VM

Hello,

In my model I would like to transpose the output right after the final convolutional layer. It has the shape of BCHW and I would like the transpose it as BHWC.

Unfortunately the TIDL is not supporting the Translate op, but I found that it has a TIDL_ShuffleChannelLayer, described as Reshape + Transpose + Reshape ops in ONNX.

I tried a simple model in torch using a single Conv2d layer and  Reshape+Permute+Reshape ops which result exactly with Reshape + Transpose + Reshape ops in ONNX. However the artifacts creation in TIDL still

get the following:

Supported TIDL layer type --- Conv -- Conv_0
Supported TIDL layer type --- Reshape -- Reshape_2
Unsupported (TIDL check) TIDL layer type --- Transpose
Supported TIDL layer type --- Reshape -- Reshape_5

Could you please explain what should be done to exploit the TIDL_ShuffleChannelLayer or suggest another way to perform a Transpose op in TIDL.

Thank you,

Alex.