TensorRT version: 7.2.2-1
CUDA version: 11.1
CUDNN version: 8.0.5
Installation instructions:
You can follow the approprivate version TLT-converter or TAO-converter (TLT-upgraded) for Hardware: TLT-converter or TAO-converter
Check version TensorRT: dpkg -l | grep TensorRT
-
Copy the executable to the target device
-
Install openssl library using the following commam.
sudo apt-get install libssl-dev
-
Export the following environment variables
export TRT_LIB_PATH=”/usr/lib/x86_64-linux-gnu"
export TRT_INC_PATH=”/usr/include/x86_64-linux-gnu”
-
Run the converter following the below guide: Sample usage
Note:
In Int8 mode, when using calibration table generated from the TLT v1.0 GA docker, if the version of the tensorrt on the deployment device is 5.x/6.x, please make sure to update the version number in the TensorRT calibration table file. Inorder to do so,
- Open the
calibration.bin
orcalibration.txt
file as generated bytlt-export
using a simple text editor - Update the first line in calibration file from
TRT-5105-EntropyCalibration2
orTRT-6001-EntropyCalibration2
toTRT-7000-EntropyCalibration2
.
You need change the first line TRT version siutable for enviroment TRT Hardware eg: TRT-7221-EntropyCalibration2 siutable for TensorRT-version
- Save the edited file with the new changes.
- Use this file with
tlt-converter
to generate the engine. - Running TAO-converter as same as TLT-coverter.
Sample usage:
Inital running:
$chmod +x tlt-converter
Sample:
$./tlt-converter resnet34_peoplenet_pruned.etlt -k tlt_encode -c resnet34_peoplenet_int8.txt -o output_cov/Sigmoid,output_bbox/BiasAdd -d channel,height,width -i nchw -e peoplenet34_int8.engine -m $MAX_BATCH_SIZE -t $INFERENCE_PRECISION -b $BATCH_SIZE
Example:
$./tlt-converter peoplenet_v2/resnet34_peoplenet_pruned.etlt -k tlt_encode -c peoplenet_v2/resnet34_peoplenet_int8.txt -o output_cov/Sigmoid,output_bbox/BiasAdd -d 3,544,960 -i nchw -e peoplenet_resnet34_int8.engine -m 1 -t int8
tlt_encode = KEY-TLT
After achived file engine, you can check inference:
Type: detectnet_v2,maskrcnn,unet Model-arch
Sample:
$tlt-infer type -e peoplenet_v2/infer_spec.txt -k tlt_encode -o infer_result -i file-image -v