Building Video AI Applications on the Edge with Jetson Nano - fp16 Engine

Hello,

I’m following the jupyter tutorial on the Video AI applications course.

I’m on section 1.2 Configure an Object Detection model where I’ve changed the model-engine-file to FP16 rather than INT8 in the config file. However when I run the deepstream_test1_rtsp_out.py again, it still builds the engine :

infer_context_impl.cpp:1947> [UID = 1]: serialize cuda engine to file: /opt/nvidia/deepstream/deepstream-6.0/samples/models/Primary_Detector/resnet10.caffemodel_b1_gpu0_fp16.engine successfully*

Does anyone know why this still happens even though I specified the file path in the config file?

Thanks.

Hi,

Do you also modify the engine file name into the fp16?

The building process will be triggered automatically if the engine doesn’t exist.
Please remember to update the file name to the corresponding engine file name and try it again.

Thanks.

Hello,

Yes I did, but it when I run the program again, it seems to read that directory but then say “failed” and then the next line goes about building the fp16 engine again.

Thanks.

Hi,

Could you share the log with us?
More, please also double-confirm if the engine exists.

Thanks.

Hi,

Apologies, the engine does exist, it’s an error on my part. Please close the thread.

Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.