Part Number: TDA2EX17EVM
Hi.
I found the TIDL translator has limitation regarding deconvolution.
I have to configure "group size" as the same value with input/output channel number.
Since the original model's group size is 1 which means each channel has it's kernel while "group size == input/output channel number" means all channel share one kernel.
I wonder this may decrease output accuracy.
Is there any method to use group size == 1?
Thanks in advance.