Deploying to Deepstream
To deploy a TAO-trained Segformer model to DeepStream, you need to use tao-converter
to
generate a device-specific optimized TensorRT engine, which can then be ingested by DeepStream.
Download the corresponding device specific tao-converter
from the
TAO converter matrix.
Machine-specific optimizations are performed as part of the engine creation process, so you should generate a distinct engine for each environment and hardware configuration. Furthermore, if the TensorRT or CUDA libraries of the inference environment are updated (including minor version updates), or if a new model is generated, you will need to generate new engines. Running an engine that was generated with a different version of TensorRT and CUDA is not supported and will cause unknown behavior that affects inference speed, accuracy, and stability–or it may fail to run altogether.
See the Exporting the Model documentation for UNet for more details on how to export a TAO model.
Segformer models require the TensorRT OSS build because several prerequisite TensorRT plugins are only available in the TensorRT open source repo.
If your deployment platform is an x86 PC with an NVIDIA GPU, follow the TensorRT OSS on x86 instructions; if your deployment platform is NVIDIA Jetson, follow the TensorRT OSS on Jetson (ARM64) instructions.
TensorRT OSS on x86
Building TensorRT OSS on x86:
Install Cmake (>=3.13).
NoteTensorRT OSS requires cmake >= v3.13, so install cmake 3.13 if your cmake version is lower than 3.13c
sudo apt remove --purge --auto-remove cmake wget https://github.com/Kitware/CMake/releases/download/v3.13.5/cmake-3.13.5.tar.gz tar xvf cmake-3.13.5.tar.gz cd cmake-3.13.5/ ./configure make -j$(nproc) sudo make install sudo ln -s /usr/local/bin/cmake /usr/bin/cmake
Get GPU architecture. The
GPU_ARCHS
value can be retrieved by thedeviceQuery
CUDA sample:cd /usr/local/cuda/samples/1_Utilities/deviceQuery sudo make ./deviceQuery
If the
/usr/local/cuda/samples
doesn’t exist in your system, you could downloaddeviceQuery.cpp
from this GitHub repo. Compile and rundeviceQuery
.nvcc deviceQuery.cpp -o deviceQuery ./deviceQuery
This command will output something like this, which indicates the
GPU_ARCHS
is75
based on CUDA Capability major/minor version.Detected 2 CUDA Capable device(s) Device 0: "Tesla T4" CUDA Driver Version / Runtime Version 10.2 / 10.2 CUDA Capability Major/Minor version number: 7.5
Build TensorRT OSS:
git clone -b 21.08 https://github.com/nvidia/TensorRT cd TensorRT/ git submodule update --init --recursive export TRT_SOURCE=`pwd` cd $TRT_SOURCE mkdir -p build && cd build
NoteMake sure your
GPU_ARCHS
from step 2 is in TensorRT OSSCMakeLists.txt
. If GPU_ARCHS is not in TensorRT OSSCMakeLists.txt
, add-DGPU_ARCHS=<VER>
as below, where<VER>
representsGPU_ARCHS
from step 2./usr/local/bin/cmake .. -DGPU_ARCHS=xy -DTRT_LIB_DIR=/usr/lib/x86_64-linux-gnu/ -DCMAKE_C_COMPILER=/usr/bin/gcc -DTRT_BIN_DIR=`pwd`/out make nvinfer_plugin -j$(nproc)
After building ends successfully,
libnvinfer_plugin.so*
will be generated under\`pwd\`/out/.
Replace the original
libnvinfer_plugin.so*
:sudo mv /usr/lib/x86_64-linux-gnu/libnvinfer_plugin.so.8.x.y ${HOME}/libnvinfer_plugin.so.8.x.y.bak // backup original libnvinfer_plugin.so.x.y sudo cp $TRT_SOURCE/`pwd`/out/libnvinfer_plugin.so.8.m.n /usr/lib/x86_64-linux-gnu/libnvinfer_plugin.so.8.x.y sudo ldconfig
TensorRT OSS on Jetson (ARM64)
Install Cmake (>=3.13)
NoteTensorRT OSS requires cmake >= v3.13, while the default cmake on Jetson/Ubuntu 18.04 is cmake 3.10.2.
Upgrade TensorRT OSS using:
sudo apt remove --purge --auto-remove cmake wget https://github.com/Kitware/CMake/releases/download/v3.13.5/cmake-3.13.5.tar.gz tar xvf cmake-3.13.5.tar.gz cd cmake-3.13.5/ ./configure make -j$(nproc) sudo make install sudo ln -s /usr/local/bin/cmake /usr/bin/cmake
Get GPU architecture based on your platform. The
GPU_ARCHS
for different Jetson platform are given in the following table.Jetson Platform
GPU_ARCHS
Nano/Tx1
53
Tx2
62
AGX Xavier/Xavier NX
72
Build TensorRT OSS:
git clone -b 21.03 https://github.com/nvidia/TensorRT cd TensorRT/ git submodule update --init --recursive export TRT_SOURCE=`pwd` cd $TRT_SOURCE mkdir -p build && cd build
NoteThe
-DGPU_ARCHS=72
below is for Xavier or NX, for other Jetson platform, change72
referring toGPU_ARCHS
from step 2./usr/local/bin/cmake .. -DGPU_ARCHS=72 -DTRT_LIB_DIR=/usr/lib/aarch64-linux-gnu/ -DCMAKE_C_COMPILER=/usr/bin/gcc -DTRT_BIN_DIR=`pwd`/out make nvinfer_plugin -j$(nproc)
After building ends successfully,
libnvinfer_plugin.so*
will be generated under‘pwd’/out/.
Replace
"libnvinfer_plugin.so*"
with the newly generated.sudo mv /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.8.x.y ${HOME}/libnvinfer_plugin.so.8.x.y.bak // backup original libnvinfer_plugin.so.x.y sudo cp `pwd`/out/libnvinfer_plugin.so.8.m.n /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.8.x.y sudo ldconfig
The tao-converter
tool is provided with the TAO Toolkit
to facilitate the deployment of TAO trained models on TensorRT and/or Deepstream.
This section elaborates on how to generate a TensorRT engine using tao-converter
.
For deployment platforms with an x86-based CPU and discrete GPUs, the tao-converter
is distributed within the TAO docker. Therefore, we suggest using the docker to generate
the engine. However, this requires that the user adhere to the same minor version of
TensorRT as distributed with the docker. The TAO docker includes TensorRT version 8.0.
Instructions for x86
For an x86 platform with discrete GPUs, the default TAO package includes the tao-converter
built for TensorRT 8.2.5.1 with CUDA 11.4 and CUDNN 8.2. However, for any other version of CUDA and
TensorRT, please refer to the overview section for download. Once the
tao-converter
is downloaded, follow the instructions below to generate a TensorRT engine.
Unzip the zip file on the target machine.
Install the OpenSSL package using the command:
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
tao-converter
using the sample command below and generate the engine.Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on x86 section above or in this GitHub repo.
Make sure to follow the output node names as mentioned in Exporting the Model
section of the respective model.
Instructions for Jetson
For the Jetson platform, the tao-converter
is available to download in the NVIDIA developer zone. You may choose
the version you wish to download as listed in the overview section.
Once the tao-converter
is downloaded, please follow the instructions below to generate a
TensorRT engine.
Unzip the zip file on the target machine.
Install the OpenSSL package using the command:
sudo apt-get install libssl-dev
Export the following environment variables:
$ export TRT_LIB_PATH=”/usr/lib/aarch64-linux-gnu”
$ export TRT_INC_PATH=”/usr/include/aarch64-linux-gnu”
For Jetson devices, TensorRT comes pre-installed with Jetpack. If you are using older JetPack, upgrade to JetPack-5.0DP.
Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on Jetson (ARM64) section above or in this GitHub repo.
Run the
tao-converter
using the sample command below and generate the engine.
Make sure to follow the output node names as mentioned in Exporting the Model
section of the respective model.
Using the tao-converter
tao-converter [-h] -k <encryption_key>
-p <optimization_profiles>
[-d <input_dimensions>]
[-o <comma separated output nodes>]
[-c </path/to/calibration/cache_file>]
[-e </path/to/output/engine>]
[-b <calibration batch size>]
[-m <maximum batch size of the TRT engine>]
[-t <engine datatype>]
[-w <maximum workspace size of the TRT Engine>]
[-i <input dimension ordering>]
[-s]
[-u <DLA_core>]
input_file
Required Arguments
input_file
: The path to the.etlt
model exported usingexport
.-p
: Optimization profiles for.etlt
models with dynamic shape. Use a comma-separated list of optimization profile shapes in the format<input_name>,<min_shape>,<opt_shape>,<max_shape>
, where each shape has the format:<n>x<c>x<h>x<w>
. This can be specified multiple times if there are multiple input tensors for the model.-k
: The key used to encode the.tlt
model when doing the traning
Optional Arguments
-e
: The path to save the engine to. The default path is./saved.engine
. Use.engine
or.trt
as an extension for the engine path.-t
: The desired engine data type. This option generates a calibration cache if in INT8 mode. The default value isfp32
. The options arefp32
,fp16
, andint8
.-w
: The maximum workspace size for the TensorRT engine. The default value is1073741824(1<<30)
.-i
: The input dimension ordering. The default value isnchw
. The options arenchw
,nhwc
, andnc
. For Segformer, you can omit this argument.-s
: A Boolean value specifying whether to apply TensorRT strict type constraints when building the TensorRT engine.-u
: Specifies the DLA core index when building the TensorRT engine on Jetson devices.-d
: A comma-separated list of input dimensions that should match the dimensions used forexport
.-o
: A comma-separated list of output blob names that should match the output configuration used forexport
.
INT8 Mode Arguments
-c
: The path to the calibration cache file for INT8 mode. The default path is./cal.bin
.-b
: The batch size used during theexport
step for INT8 calibration cache generation (default:8
).-m
: The maximum batch size for the TensorRT engine. The default value is16
. If you encounter out-of-memory issues, decrease the batch size accordingly. This parameter is not required for.etlt
models generated with dynamic shape (which is only possible for new models introduced in TAO Toolkit 3.21.08).
Sample Output Log
Here is a sample log for exporting a Segformer model.
tao-converter -k $KEY
-c $USER_EXPERIMENT_DIR/export/isbi_cal.bin
-t fp16
-p input,1x1x512x512,4x1x512x572,16x1x512x512
/path/to/segformer.etlt
..
[INFO] Some tactics do not have sufficient workspace memory to run. Increasing workspace size may increase performance, please check verbose output.
[INFO] Detected 1 inputs and 2 output network tensors.
To use the default tao-converter
available in the TAO Toolkit
package, append tao
to the sample usage of the tao_converter
.
Once the model and/or TensorRT engine file have been generated, two additional files are required:
The label file is a text file containing the names of the classes that the Segformer model
is trained to segment. The order in which the classes are listed here must match the order
in which the model predicts the output. This order is derived from the
target_class_id_mapping.json
file that is saved in the results
directory after
training. Here is an example of the target_class_id_mapping.json
file:
{"0": ["foreground"], "1": ["background"]}
Here is an example of the corresponding segformer_labels.txt
file. The order in the
segformer_labels.txt
should match the order of the target_class_id_mapping.json
keys:
foreground
background
The segmentation model is typically used as a primary inference engine. It can also be used as a
secondary inference engine. Download ds-tlt
from the deepstream_tao_apps repo.
Follow these steps to use the TensorRT engine file with the ds-tlt
:
1. Generate the TensorRT engine using tao-converter
. Detailed instructions are provided in
the Generating an engine using tao-converter
section.
Once the engine file is generated successfully, do the following to set up ds-tlt with DS 6.1.
Follow the instructions here to install ds-tlt: DS TAO installation.
To run this model with the sample ds-tao-segmentation
, you must modify
the existing pgi_segformer_tlt_config.txt
file here to point to this model.
For all options, see the configuration file below. To learn more about the parameters, refer to the
DeepStream Development Guide.
[property]
gpu-id=0
net-scale-factor=0.007843
# 0-RGB, 1-BGR, 2-Gray
model-color-format=1 # For grayscale, this should be set to 2
offsets=127.5; 127.5; 127.5
labelfile-path=</Path/to/segformer_labels.txt>
##Replace following path to your model file
model-engine-file=<Path/to/tensorrt engine generated by tao-converter>
#current DS cannot parse segformer etlt model, so you need to
#convert the etlt model to TensoRT engine first use tao-converter
infer-dims=c;h;w # where c = number of channels, h = height of the model input, w = width of model input.
batch-size=1
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=2
num-detected-classes=2
interval=0
gie-unique-id=1
## 0=Detector, 1=Classifier, 2=Semantic Segmentation (sigmoid activation), 3=Instance Segmentation, 100=skip nvinfer postprocessing
network-type=100 # set this to 2 if sigmoid activation was used for semantic segmentation
output-tensor-meta=1 # Set this to 1 when network-type is 100
output-blob-names=argmax_1 # If you had used softmax for segmentation model, it would have beedn replaced with argmax by TAO for optimization. Hence, you need to provide argmax_1
segmentation-threshold=0.0
##specify the output tensor order, 0(default value) for CHW and 1 for HWC
segmentation-output-order=1
[class-attrs-all]
roi-top-offset=0
roi-bottom-offset=0
detected-min-w=0
detected-min-h=0
detected-max-w=0
detected-max-h=0
The following is an example of a modified config file for a resnet18
3-channel model trained on the ISBI dataset:
[property]
gpu-id=0
net-scale-factor=0.007843
# Since the model input channel is 3, and pre-processing of SegFormer TAO requires BGR format, set the color format to BGR.
# 0-RGB, 1-BGR, 2-Gray
model-color-format=1 # For grayscale, this should be set to 2
offsets=127.5;127.5;127.5
labelfile-path=/home/nvidia/deepstream_tlt_apps/configs/segformer_tlt/segformer_labels.txt
##Replace following path to your model file
# Argument to be used if you are using an tensorrt engine
model-engine-file=/home/nvidia/deepstream_tlt_apps/models/segformer/segformer_isbi.engine
infer-dims=3;512;512
batch-size=1
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=2
num-detected-classes=2
interval=0
gie-unique-id=1
## 0=Detector, 1=Classifier, 2=Semantic Segmentation (sigmoid activation), 3=Instance Segmentation, 100=skip nvinfer postprocessing
network-type=100
output-tensor-meta=1 # Set this to 1 when network-type is 100
output-blob-names=argmax_1 # If you had used softmax for segmentation model, it would have been replaced with argmax by TAO for optimization.
# Hence, you need to provide argmax_1
segmentation-threshold=0.0
##specify the output tensor order, 0(default value) for CHW and 1 for HWC
segmentation-output-order=1
[class-attrs-all]
roi-top-offset=0
roi-bottom-offset=0
detected-min-w=0
detected-min-h=0
detected-max-w=0
detected-max-h=0
Currently, Segformer only supports TensorRT Engine input in the DS configuration file.
Convert the .etlt
engine to .trt
using tao-converter.
Below is a sample ds-tao-segmentation
command for inference on a single image:
ds-tao-segmentation -c pgie_config_file -i image_isbi_rgb.jpg