Deepstream-occupancy-analytics display RTSP freezes and blurs on jetson nano

• Hardware Platform (Jetson nano)
• DeepStream Version 5.0.1
• JetPack Version 4.4
• TensorRT Version 7
• Issue Type(bug)

My software is the latest. Use a stable 5V 4A power supply, adjust to the maximum performance mode
When running deepstream-occupancy-analytics, the video stream will freeze, blur, and skip frames.
I tried two brands of cameras, and tried all the settings related to the video stream (H264/H265, resolution, key frame interval, bit rate).
And everything works fine when running with deepstream-test3-app (video here)

I recorded a video of the deepstream-occupancy-analytics running
with jtop video

The configuration file I tried for the first time is as follows:
test5_config_file_src_infer_tlt.txt (4.6 KB)
Log output:
bug.txt (511.1 KB)

After I closed the primary-gie and tracker in the configuration file, the display screen started to freeze at intervals

By the way,when I was running deepstream-test5 (using the original configuration file), the screen started to be smooth and then it started to freeze(video here)

The default config may be too heavy for Jetson Nano. Please set type=5(nvoverlaysink) in [sink0] and execute sudo jetson_clocks.

I have tried setting the type to 5 a long time ago, and the app directly reported an error and failed to run. I also tested on xavier. One video runs normally, but two videos start to freeze. Can you try it yourself, I have two Different versions of jetson nano and jetson xavier, none of them can work normally

I used deepstream-test5 to test again. A small model was used in deepstream-test5. I think jetson nano is fully capable of running smoothly. Change the default type=2 (running stuck) to type=5, I get

The modified configuration file is as follows
test5_config_file_src_infer.txt (4.6 KB)

We have checked and the model you are using is overload for Nano platform. You would need to set the appropriate interval property in nvinfer section of the config file to get real time performance. And may try to use Resnet18 model available on NGC with non zero interval. Please refer to performance guide in document: