VPI: Unable to write to VPI_ARRAY_TYPE_U8 on jetson device

Hi @AastaLLL

Any progress? Hope you all had a good easter break!

Kind regards,

Tom.

Hi @AastaLLL

I guess this has gone off your radar a bit. Could you give me an update?

Thanks,

Tom.

Hi,

Sorry to keep you waiting.

Our internal team is still working on it.
Will share more information with you later.

Thanks.

Hi @AastaLLL

I hope you are well. I spied that there is a small new release of jetpack (5.1.1) and wondered if there were any changes that could help my case.

Kind regards,

Tom.

Hi,

Unfortunately, the issue is not fixed yet.
But you can still try it with the latest VPI library.

Thanks.

Hi AastaLLL, Any update?

Hi,

Thanks for getting in touch with us.
Unfortunately, our internal team is still working on this.

Thanks.

Hi Aasta, another 3 months have gone by. Could you please give me an update on this issue? Despite the time period, I am still keen that this is fixed. Could you give me any kind of duration on when this will be fixed and if there is actually anyone working on this?

I note that there is a new version of VPI (2.3) but sadly see no note of my issue in the list of known issues.

I have recently retried this without the OpenCV wrapper functions (instead using VPI images directly) and can confirm this issue is still present on a Jetson Xavier NX using jetpack 5.1.

Kind regards,

Tom.

Hi,

Thanks for your patience and sorry for keeping you waiting.

We do have an internal bug to track this but it takes time due to the limited resources.
For now, please use the WAR (recreating the VPIArrray) instead.

We will let you know once we get any further updates.

Has the problem been solved? I want to use vpiSubmitOpticalFlowPyrLK to track feature points continuously. When I use vpiSubmitOpticalFlowPyrLK (stream, 0, optflow pyrPrevFrame, pyrCurFrame, prevFeatures, curFeatures, status, &lkParams), the status cannot be updated normally, and the data in status can only be changed from 0 to 1, but not from 1 to 0. I think it’s a bug.

Hi,

Have you checked if you are facing the same issue as this topic?
If yes, could you try the workaround which is to recreate the VPIArray?

Although it will cause some performance overhead, it can unblock the issue and move on.

Thanks.

Hi, @tom-fred

We got some updates from our internal team.

Our internal team has located the possible root of the problem but it might take some time to fix.
Moreover, if you try to replicate the OpticalPyrLK sample, you don’t need to reset the status array to 0 after every iteration.

Tracking from one frame to another does not require that all statuses be set to 0.
It needs to be set to 0 only if there is a different video that needs to be tracked.

Thanks.

Hi Aasta, thanks for the update.

It will be interesting to see how things develop. Will this be affected by VPI 3.0 coming out and Jetpack 6.0? (I am thinking of the Orin-only nature of this release.

Thanks.

Hi,

Our internal team is still working on it.
Unfortunately, we don’t have an ETA can share here.

Thanks.