Better solution on tensor.unsqueeze function available

[robberpen]

Hi Team

Based on #3777

My voice model input as tensor float32[1,16000,1], but element alsasrc is [1,16000]
Is any tensor unsqueeze feature in nnstreamer? Or what is better solution such as customize unsqueeze feature on tensor_filter/tensor_transform element?

[anyj0527]

It the pipeline does not work, you may try adding other/tensors,format=static,num_tensors=1,types=float32,dimensions=1:16000:1 before tensor_filter.

[myungjoo]

You don't need to unsqueeze it. N:M is dealt equivalent to N:M:1. However, if you want to unsqueeze 1:N:M to N:M, use tensor_transform.

[robberpen]

Thanks your replay, and yes, it don't need to unsqueeze it, and working in my case now.
However, an error occurred on if I connect to tensor_decoder mode=image_labeling which it accept 2dim.
I just modified tensordec-imagelabel.c below and working now, No idea any side effect.

tensordec-imagelabel.c

il_setOption()
   g_return_val_if_fail (dim[0] > 0 && dim[1] == 1, NULL);
-  for (i = 2; i < NNS_TENSOR_RANK_LIMIT; i++)
+  for (i = 3; i < NNS_TENSOR_RANK_LIMIT; i++)
     g_return_val_if_fail (dim[i] == 0, NULL);

[myungjoo]

It appears that it needs to be updated like this:

-  g_return_val_if_fail (dim[i] == 0, NULL);
+  g_return_val_if_fail (dim[i] == 0 || dim[i] == 1, NULL);

@niley7464 Please take a look at this.