Converting tao trained model to deepstream engine

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Using this command,

mask_rcnn export -m /workspace/Nyan/cv_samples_v1.3.0/mask_rcnn/experiment_dir_retrain/model.step-90000.tlt -k nvidia_tlt -e /workspace/Nyan/cv_samples_v1.3.0/mask_rcnn/specs/maskrcnn_retrain_resnet18.txt --batch_size 1 --data_type int8 --cal_image_dir /workspace/Nyan/cv_samples_v1.3.0/mask_rcnn/images --batches 10 --cal_cache_file /workspace/Nyan/cv_samples_v1.3.0/mask_rcnn/export/maskrcnn.cal --cal_data_file /workspace/Nyan/cv_samples_v1.3.0/mask_rcnn/export/maskrcnn.tensorfile

I try to produce etlt file and cal.bin file so that I can create engine file in Jetson device. But what I have is etlt file, cal file and tensorfile.

Using deepstream, I need etlt file and bin file, how to make engine file in jetson?

3

See MaskRCNN — TAO Toolkit 3.22.05 documentation
After exporting, you will have .etlt file and cal.bin file.

Then, you can use tao-converter to generate tensorrt engine in Jetson.

4

Yes thanks.
maskrcnn.cal file is .bin file.
At Xavier, when I do conversion for engine file, the error is

UffParser: Output error: Output mask_head/mask_fcn_logits/BiasAdd not found

In MaskRCNN — TAO Toolkit 3.22.05 documentation,

tao mask_rcnn export produces only etlt file and bin file. I copied etlt file and bin file to Xavier and do engine conversion.

Export step doesn’t fixed output nodes and only ‘tao-converter’ setup output nodes.

5

Refer to MaskRCNN — TAO Toolkit 3.22.05 documentation

6

Yes I follow option 1

  • Option 1: Integrate the model (.etlt) with the encrypted key directly in the DeepStream app. The model file is generated by export.

etlt file and cal file are specified in config file as

tlt-encoded-model=../../../../samples/models/maskrcnn/maskrcnn.etlt
tlt-model-key=nvidia_tlt
model-engine-file=../../../../samples/models/maskrcnn/maskrcnn.etlt_b1_gpu0_int8.engine
int8-calib-file=../../../../samples/models/maskrcnn/maskrcnn.cal

When generate engine, why I have this error
ERROR: [TRT]: UffParser: Output error: Output mask_head/mask_fcn_logits/BiasAdd not found
Why my maskrcnn.etlt doesn’t have those outputs?

I did follow all steps for

TensorRT OSS 
libnvds_infercustomparser_tao.so
7

Found out that I don’t need mask_head

8

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