TDA4VM: The problem of TIDL

ZHIJIE PAN

Part Number: TDA4VM

Hi experts,

I have a question. When the model is imported for TIDL, this function will check the dimension of the input tensor, which needs to be greater than 0 in all four dimensions. If for one-dimensional input, does it need to be setted as 1 x 1 x 1 x N ?

In addition, does the Concat layer support splicing three one-dimensional inputs？

Bes regards,

Zhijie

over 3 years ago

0 Anshu Jain over 3 years ago

TI__Guru 56820 points

Hi Zhijie,

ZHIJIE PAN said:
If for one-dimensional input, does it need to be setted as 1 x 1 x 1 x N ?

Yes

ZHIJIE PAN said:
does the Concat layer support splicing three one-dimensional inputs

Here do you mean concat operation across different access?

Regards,
Anshu

0 ZHIJIE PAN over 3 years ago in reply to Anshu Jain

Prodigy 90 points

Hi Anshu,

Thanks for replying. What I mean is that does Concat layer support 3 different inputs from different access? Also, the input is the 1x1x1xN vector. So, the Concat layer needs to be set as 1xNx1x1, right?(N is the channel)

Best regards,

Zhijie

0 Anshu Jain over 3 years ago in reply to ZHIJIE PAN

TI__Guru 56820 points

Hi Zhijie,

I am still not following your question, can you give an example of what you are looking for ?

Regards,
Anshu

0 ZHIJIE PAN over 3 years ago in reply to Anshu Jain

Prodigy 90 points

Hi Anshu,

This is the example.

0 Anshu Jain over 3 years ago in reply to ZHIJIE PAN

TI__Guru 56820 points

Hi Zhijie,

This is supported by TIDL, assuming what you wrote is in NCHW format in this case concatenate will be happening across width.

Regards,

Anshu

Processors

Processors forum

TDA4VM: The problem of TIDL