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: TIDL-concat and MIPI have problems working at the same time

Part Number: TDA4VM


Hi All

SDK: 7.2

kernel :rootfs

I run the following TIDL model witch model contains a Concat layer, the layer output size is 1x512x45x80,at the same time, run 8-channel camera capture,

Will most likely cause the view capture error。

  • Hi shuyu liang,

    Kindly share more details about the issue, like what is use case that you are running and what is the issue you are facing with concat layer  ?

    Thanks,

    Praveen

  • Hi Praveen

    The case implementation is roughly as follows :

    When I collected an 8-channel image and ran a backbone model with tidl, I found that the image was abnormal.

    I decomposed the model layer by layer and finally confirmed that it was the Concat problem.

    The output size of this Concat is 1x512x45x80.

    Tests that have been found to not cause problems are as follows:

    1. Collect 1-channel images and run the model without problems

    2. Collect 8-channel images, change the Concat output size to 1x256x45x80, there will be no problem

    3. Collect 8-channel images, change the Concat output size to 1x512x30x80, there will be no problem

    Increase in the number of acquisition channels(maybe  5 、 6 ...), or an increase in the size of the Concat output can cause problems.

  • Hi, 

    Is it possible for you to share a model with Increase in the number of acquisition channels (maybe  5 , 6 ...)to replicate the issue at our end in the Concat layer ?

    Thanks,

    Praveen

  • Hi,

    This issue is being discussed over the email with Joe(joe.shen@ti.com) so closing this ticket. 

    Regards,

    Brijesh