Object detection and Counts Decrease issue + MIPI CSI 2 + TX2

We have prepared a project which detects a particular object and uniquely keeps a count of it through out the video-feed. We have a setup where the camera is facing the laptop which plays a video on loop, the problem that we’re facing is, after a certain duration the count of the object keeps on decreases at a random interval. Lets say initially its around 300 after couple of days its around 100 and as mentioned above it is detecting and counting the same object with the same looped video.

The project has Jetson TX2 with MIPI CSI Camera along with latest Jetpack and TensorRT.

Is there any issue with the hardware’s used?
Is FPS a key aspect? its around 10. Is it ok?
Does anything in our setup is causing the above mentioned issue?
Will FPS count decrease after certain period of time due to hardware limitation ? Is there any such possibility?

Thanks in advance.

Is there any issue with the hardware’s used? I don’t think so.
Is FPS a key aspect? its around 10. Is it ok? It depends on the model you choose, e.g yolov3, it makes sense to have 10 fps on TX2.
Does anything in our setup is causing the above mentioned issue? I guess no.
Will FPS count decrease after certain period of time due to hardware limitation ? Is there any such possibility? The post process will cause difference, but you should check the CPU/GPU/thermal status when the issue happen.

Could you please help me understand why am i observing a decrease in count for the same looped video?. As Mentioned in my above description.

Sorry, the information is not enough.
And I don’t think we can get any useful information from it.

I would suggest to do some engineering work to get more data/clue.
Also check your code as well.
This will help the others to help.

Could you please elaborate on what sort of more data / clue would help out in solving this issue?

I think we are facing a similar issue as the one mentioned in MIPI CSi frame rate issue ,
please share your thoughts on it.

could you help us with a code snippet to get FPS per frame?

Thanks

Some additional information, we are performing object detection and tracking using the TX2 along with MIPI CSI 2 camera and TensorRT optimized model, we have observed fluctuations in the FPS.

Could you help us understand what could be the reason for this?

Hi,

Sorry for the late update.

Not sure which frameworks are you using for camera.
If you didn’t use our DeepstreamSDK, would you mind to give it a try?

The implementation in DeepstreamSDK goes through our QA process (including long-run), so it should not have this issue.

Thanks.