Other Parts Discussed in Thread: TDA4VH
Hi TI：
The board I'm currently using is the 9.0 SDK for TDA4VH.
I have a Slice layer in my model that splits a tensor T= [1, 123, 16, 44] and takes the first 59 channels to form a new output A=[1, 59 ,16, 44], and the last 64 channels to form another output B=[1, 64, 16, 44].
（output A = T[1, :59, 16, 44] output B = T[1, 59:, 16, 44] 59+64=123）
But I found that the output A of 59 channels and the output B of 64 channels in the PC Importer trace dump layer are consistent with the inference results of the onnx model, but when I run the model exported by PC Importer on the board to get the output, I find that the output A of 59 channels is correct and the output B of 64 channels is incorrect.
So, is there a limit to the use of the Sclice layer? What causes this?
Thank you
