How to export a trt file from a fpenet model.tlt?

Hi,

I am using the TLT v3.0 docker container and need to convert the pre-trained FPENet model.tlt that I obtained by

ngc registry model download-version nvidia/tlt_fpenet:trainable_v1.0 --dest $PWD

to an engine file (.trt). But the command fpenet export accepts no argument for engine-file. I know that it is possible to get a .trt file for detectnet_v2 model.tlt.

Here is the output of the commands:

Thanks.

Can you follow Facial Landmarks Estimation — Transfer Learning Toolkit 3.0 documentation to export to etlt file?

tlt fpenet export -m <Trained TLT Model Path> -k <Encode Key> -o <Output file .etlt>

For how to generate trt engine in fpenet, please see similar topic

You can download via

ngc registry resource download-version “nvidia/tlt_cv_inference_pipeline_quick_start:v0.1-dp”

https://docs.nvidia.com/metropolis/TLT/tlt-user-guide/text/tlt_cv_inf_pipeline/requirements_and_installation.html#download-the-tlt-cv-inference-pipeline-quick-start

Then, in tlt_cv_compile.sh, you can refer to below to generate trt engine.

     tlt-converter -k ${ENCODING_KEY} -t fp16 \
            -p input_face_images:0,1x1x80x80,1x1x80x80,2x1x80x80 \
            -e /models/triton_model_repository/faciallandmarks_tlt/1/model.plan \
            /models/tlt_cv_faciallandmarks_v${tlt_jarvis_ngc_version}/faciallandmarks.etlt
1 Like

Thanks, this solved my issue.