Custom trained model Detectnet V2 and deploying to jetson nano

[Will00]

Hi, i have train a custom model detectnet_v2 lpdnet by using notebook from tao launcher starter kit and create the .tlt model. I’m using Ubuntu 20.04 WSL2 with GPU RTX3060.
This is the result from training process:

When i try to deploy using jetson nano using deepstream-app there is a error like this:

Any ideas?
Thanks.
• Hardware Platform (Jetson Nano)
• DeepStream Version 6.0.1**
• JetPack Version 4.6.1
• TensorRT Version 8.2.1
• CUDA 10.2

[yingliu]

Can you share the full log and config file?

[Will00]

Sure, the config is similar to lpdnet us
lpd_id_config.txt (1012 Bytes)

[Fiona.Chen]

Please read the log. it can’t find the etlt file. Please check whether there is /home/irumadev/Downloads/expe/deepstream_lpr_app/deepstream-lpr-app/resnet_lpd/resnet18_detector_pruned_lpd.etlt exists in your machine.

[Will00]

I already try that too and the this is the results:

Is it because .etlt file failed during exporting from tlt to etlt?

[Morganh]

@Will00

You set

tlt-encoded-model=resnet_lpd/resnet18_detector_pruned_lpd.tlt

That is not expected.
Please set it to an etlt file.

[Will00]

Sorry for late respond, i just try to redo the entire process on the notebook.
Its working on jetson nano.

[Fiona.Chen]

Glad to know it works. Any more issues?

[Will00]

Last one, Do you have any ideas sometimes why fps is dropped?

Adapter Power issue?

[Fiona.Chen]

You may refer to Troubleshooting — DeepStream 6.3 Release documentation

And you can use “nvidia-smi dmon” for x86 and “tegrastats” for Jetson to monitor the GPU/CPU usage during the app running.

[Will00]

$ sudo nvpmodel -m 0
$ sudo jetson_clocks
Already done.
But sometimes fps still drop maybe i should try change the power adapter.

[Fiona.Chen]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Have you monitored the CPU and GPU usage during the app is running?