When the project USES detection + tracking, the detection model fails

I have a human flow detection project, human body detection + track tracking, which USES. TRT tensorRT for detection and. Pbfile tf for tracking.
Failure:
I. the Load tracking
Ii. The Load detection
Iii. Infer detection
Iv. Infer tracking
Success:
I. the Load detection
Ii. Infer detction
Iii. The Load tracking
Iv. Infer teacking

The failed detection model means that it will return 725 0.25 scores and incorrect boxes, as well as a classification value of 0

May I ask why this is?

How do I solve this

Hi,

Sorry that we will need more information to give a further suggestion.

A possible cause is that most of the memory is occupied by the TensorFlow if loading the tracking model first.
And the occupation leads to some error in the inference.

Could you run the tegrastats to monitor the system status of each case and share it with us?

sudo tegrastats

Thanks.

I don’t think this command gives any clue, it just provides some resource usage, no error logs, or any other relevant information

Hi,

Based on the resource log, do you run out of memory when launching TensorFlow?

Thanks.

Both success and failure are out of memory, but the order of success is fine.
I’ll come back later and post more information.

Hi gzchenjiajun,

Any result can be shared?

This problem has not been solved, you can see here, I need your help, thank you

"Cuda Error in NCHWTONCHHW2: 33 (invalid resource handle) ",How to solve it? - NVIDIA Developer Forums
https://devtalk.nvidia.com/default/topic/1072765/jetson-nano/-quot-cuda-error-in-nchwtonchhw2-33-invalid-resource-handle-quot-how-to-solve-it-/post/5436531/#5436531

Let check this issue on the new topic directly.