Scaling issue on VIC vs GPU (Memory leak?)

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hello
I have a pipeline with 1 PGIE and 1 SGIE.
Originally I used scaling-compute-hw as VIC for SGIE but I got this error

 VIC Configuration failed image scale factor exceeds 16, use GPU for Transformation

So based on other posts in the forums, one option was to change the scaling hw to GPU.
After doing that, I get another error

/dvs/git/dirty/git-master_linux/nvutils/nvbufsurftransform/nvbufsurftransform.cpp(3446) : getLastCudaError() CUDA error : Recevied NvBufSurfTransformError_Execution_Error : (2) out of memory.

Note, this happens at random intervals (sometimes after a 5 hour run and sometimes after 2 hours). I see the RAM usage increasing until.

This happens on both Deepstream 6.0.1 and 6.1.1.
Is there any way to fix this? My SGIE’s input size is 96 x 96.
Note, I cannot upgrade Deepstream or Jetpack on either of these machines currently.

Thanks.

• Hardware Platform (Jetson / GPU): Jetson Xavier AGX and ORIN
• DeepStream Version: 6.0.1 & 6.1.1
• JetPack Version (valid for Jetson only): 4.6.1 and 5.0.1
• Issue Type( questions, new requirements, bugs): Bug

3

Could you refer to the link below and have a try?
https://forums.developer.nvidia.com/t/deepstream-sdk-faq/80236/29

4

I have tried that setting and it still crashes

input-object-min-width=6
input-object-min-height=6

But I think I am hitting a max width or height limitation
When I set the max to this, it works well.

input-object-max-width=1000
input-object-max-height=1000

Is there any correct max value with respect to input size? I thought it would be model input width * 16 x model input height * 16 but it still gives the scaling error for those values.

With respect to my second point about switching to GPU, is there any information available on how to debug/fix that?

Thanks!

5
  1. We sugget min-width > width/16, max-width < width*16 min-height> height/16, max-height< height*16.
  2. You can refer to the link below to check if there is memory leak.
    https://forums.developer.nvidia.com/t/deepstream-sdk-faq/80236/14
6
  1. I tried with that setting but seem any value greater than model width * 11 and model height * 11 fails.
  2. Thank you, I will give this a try and update.
7

Hi @marmikshah ,
May I know if there is any update from your side? Kindly let us know if this topic can be closed or not.

8

Hi,
Sorry for the late reply. You can close this one for now as I may need some more time to study the memory leak problem.
Will open a new thread if needed.
Thanks.

9