Jetson Nano YOLOv5 TensorRT DeepStream Setup Guide

Summarize with AI: ChatGPT Perplexity Claude Grok

Introduction

Train your own YOLOv5 model on a CUDA-capable host PC, convert it to a TensorRT engine, deploy it to a Jetson Nano, and run it with DeepStream. (This note uses <code>yolov5s.pt</code> as an example.)

Environment

Hardware:

CUDA-capable GPU host PC
Jetson Nano 4GB B01
CSI camera, USB camera

Software:

yolov5-5.0
jetpack-4.4
deepstream-5.0
TensorRT-7.1
CUDA-10.2

file

On the Host PC

Download YOLOv5

git clone -b v5.0 https://github.com/ultralytics/yolov5.git
git clone -b yolov5-v5.0 https://github.com/wang-xinyu/tensorrtx.git

Official tutorial documentation:

tensorrtx/yolov5 at yolov5-v5.0 · wang-xinyu/tensorrtx (github.com)

If you can't serialize an engine on the PC, you can also do it on the Jetson.

On the Jetson

Set up the YOLOv5 environment

git clone https://github.com/ultralytics/yolov5.git

python3 -m pip install --upgrade pip

Enter the YOLOv5 project directory.

pip3 install -r requirements.txt

If you hit a Pillow-related error: uninstall Pillow with <code>pip3 uninstall pillow</code>, then reinstall with <code>pip3 install pillow</code>.

Clone TensorRTx

Repository: https://github.com/wang-xinyu/tensorrtx.git

git clone https://github.com/wang-xinyu/tensorrtx.git

Place the generated <code>.wts</code> file under <code>tensorrtx/yolov5/</code>.

If you trained your own model, you need to modify <code>tensorrtx/yolov5/yololayer.h</code>:

static constexpr int CLASS_NUM = 80

Compile the code:

cd {tensorrtx}/yolov5/
mkdir build
cd build
cp {ultralytics}/yolov5/yolov5s.wts {tensorrtx}/yolov5/build
cmake ..
make

Convert the generated <code>.wts</code> file into an <code>.engine</code> file:

sudo ./yolov5 -s yolov5s.wts yolov5s.engine s

file

Put the test images under <code>tensorrtx/yolov5/samples/</code> and test whether it can detect objects:

sudo ./yolov5 -d ../best.engine ../samples

After testing, inference speed improves significantly after converting to TensorRT.

Image size: 640*640

YOLOv5s inference latency is around 130 ms

file

After conversion to TensorRT, inference latency is around 70 ms

file

Install and test DeepStream (5.0)

‼️Always check the official documentation for the matching DeepStream and JetPack versions (for example, JetPack 4.6 supports DeepStream 6.0).

file

Official documentation: NVIDIA Metropolis Documentation

Official download: DeepStream Getting Started | NVIDIA Developer

Older versions: NVIDIA DeepStream SDK on Jetson (Archived) | NVIDIA Developer

Condensed notes

sudo apt install \
    libssl1.0.0 \
    libgstreamer1.0-0 \
    gstreamer1.0-tools \
    gstreamer1.0-plugins-good \
    gstreamer1.0-plugins-bad \
    gstreamer1.0-plugins-ugly \
    gstreamer1.0-libav \
    libgstrtspserver-1.0-0 \
    libjansson4=2.11-1
sudo apt-get install librdkafka1=0.11.3-1build1
tar -xpvf deepstream_sdk_v4.0.2_jetson.tbz2
cd deepstream_sdk_v4.0.2_jetson
sudo tar -xvpf binaries.tbz2 -C /
sudo ./install.sh
sudo ldconfig

Detailed version

Install and test DeepStream (there are detailed docs in the official release)

Install required packages:

sudo apt install \
libssl1.0.0 \
libgstreamer1.0-0 \
gstreamer1.0-tools \
gstreamer1.0-plugins-good \
gstreamer1.0-plugins-bad \
gstreamer1.0-plugins-ugly \
gstreamer1.0-libav \
libgstrtspserver-1.0-0 \
libjansson4=2.11-1

Download the SDK and copy it to the Jetson
Extract the SDK

sudo tar -xvf deepstream_sdk_v5.1.0_jetson.tbz2 -C /
cd /opt/nvidia/deepstream/deepstream-5.1
sudo ./install.sh
sudo ldconfig

Test after installation

cd /opt/nvidia/deepstream/deepstream-5.0/samples/configs/deepstream-app/

deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt

Install DS-6.0

Quickstart Guide — DeepStream 6.0 Release documentation (nvidia.com)

rscgg37248/DeepStream6.0_Yolov5-6.0: object detection based on DeepStream 6.0 and YOLOv5 6.0 (GitHub)

Install

sudo apt install \
libssl1.0.0 \
libgstreamer1.0-0 \
gstreamer1.0-tools \
gstreamer1.0-plugins-good \
gstreamer1.0-plugins-bad \
gstreamer1.0-plugins-ugly \
gstreamer1.0-libav \
libgstrtspserver-1.0-0 \
libjansson4=2.11-1

sudo tar -xvf deepstream_sdk_v6.0.0_jetson.tbz2 -C /
cd /opt/nvidia/deepstream/deepstream-6.0
sudo ./install.sh
sudo ldconfig

Test

cd /opt/nvidia/deepstream/deepstream-6.0/samples/configs/deepstream-app/

deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt

YOLOv5 detection

‼️Jetson Nano system version is 4.5.1, TensorRT version is 7.x, YOLOv5 version is 5.0. If you hit Pillow-related errors: uninstall Pillow with <code>pip3 uninstall pillow</code> , then reinstall with <code>pip3 install pillow</code> .

After installing DeepStream, you will find an official YOLO deployment sample under <code>/opt/nvidia/deepstream/deepstream-5.0/sources/objectDetector_Yolo</code>, but it only supports YOLOv3.

On GitHub there are projects that have already adapted YOLOv5:

DanaHan/Yolov5-in-Deepstream-5.0: Describe how to use YOLOv5 in DeepStream 5.0 (GitHub)

I used the TensorRT 7–compatible branch:

Abandon-ht/Yolov5-in-Deepstream-5.0: TensorRT 7 branch (GitHub)

Clone the project:

git clone https://github.com/DanaHan/Yolov5-in-Deepstream-5.0.git

Test:

cd &quot;Yolov5-in-Deepstream-5.0/Deepstream 5.0&quot;

# Copy COCO dataset labels
cp ~/darknet/data/coco.names ./labels.txt

# Copy the engine file generated earlier into the current directory
cp ~/tensorrtx/yolov5/build/yolov5s.engine ./

cd nvdsinfer_custom_impl_Yolo

# Build libnvdsinfer_custom_impl_Yolo.so
make -j

# Back to Deepstream 5.0/
cd ..

# Run
LD_PRELOAD=./libmyplugins.so deepstream-app -c deepstream_app_config_yoloV5.txt

file

Use CSI or USB cameras in DeepStream

Reference: How to use USB and CSI cameras in deepstream-app (Elecfans)

# Install v4l-utils
apt-get install v4l-utils

# List camera devices
v4l2-ctl --list-devices

Check available camera resolutions:

v4l2-ctl --list-formats-ext --device=0
v4l2-ctl --list-formats-ext --device=1

Modify the <code>source</code> in <code>deepstream_app_config_yoloV5.txt</code>.

I personally use a Logitech <code>c920</code> webcam (USB).

Parameters:

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
camera-width=1280
camera-height=720
camera-fps-n=30
camera-v4l2-dev-node=0
#uri=file://../../samples/streams/sample_1080p_h264.mp4
num-sources=1
gpu-id=0
# (0): memtype_device   - Memory type Device
# (1): memtype_pinned   - Memory type Host Pinned
# (2): memtype_unified  - Memory type Unified
cudadec-memtype=0

file

Plugin configuration

Refer to the config files under <code>deepstream_sdk_v4.0.2_jetson/samples/configs/deepstream-app/</code>:

source30_1080p_resnet_dec_infer_tiled_display_int8.txt: demonstrates 30-stream decode with primary inference. (dGPU and Jetson AGX Xavier platforms only.)
source4_1080p_resnet_dec_infer_tiled_display_int8.txt: demonstrates four-stream decode with primary inference, object tracking, and three different secondary classifiers. (dGPU and Jetson AGX Xavier platforms only.)
source4_1080p_resnet_dec_infer_tracker_sgie_tiled_display_int8_gpu1.txt: demonstrates four-stream decode on GPU 1 with primary inference, object tracking, and three different secondary classifiers (for systems with multiple GPU cards). dGPU platforms only.
config_infer_primary.txt: configures the <code>nvinfer</code> element as a primary detector.
config_infer_secondary_carcolor.txt, config_infer_secondary_carmake.txt, config_infer_secondary_vehicletypes.txt: configure the <code>nvinfer</code> element as secondary classifiers.
iou_config.txt: configures a low-level IOU (intersection-over-union) tracker.
source1_usb_dec_infer_resnet_int8.txt: demonstrates using a USB camera as input.
source1_csi_dec_infer_resnet_int8.txt: demonstrates using a CSI camera as input; Jetson only.
source2_csi_usb_dec_infer_resnet_int8.txt: demonstrates using one CSI camera and one USB camera as input; Jetson only.
source6_csi_dec_infer_resnet_int8.txt: demonstrates using six CSI cameras as input; Jetson only.
source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt: demonstrates 8x decode + inference + tracker; Jetson Nano only.
source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_tx1.txt: demonstrates 8x decode + inference + tracker; Jetson TX1 only.
source12_1080p_dec_infer-resnet_tracker_tiled_display_fp16_tx2.txt: demonstrates 12x decode + inference + tracker; Jetson TX2 only.

Video input

Default test video

[application]
enable-perf-measurement=1
perf-measurement-interval-sec=5
#gie-kitti-output-dir=streamscl

[tiled-display]
enable=0
rows=1
columns=1
width=1280
height=720
gpu-id=0
#(0): nvbuf-mem-default - Default memory allocated, specific to particular platform
#(1): nvbuf-mem-cuda-pinned - Allocate Pinned/Host cuda memory, applicable for Tesla
#(2): nvbuf-mem-cuda-device - Allocate Device cuda memory, applicable for Tesla
#(3): nvbuf-mem-cuda-unified - Allocate Unified cuda memory, applicable for Tesla
#(4): nvbuf-mem-surface-array - Allocate Surface Array memory, applicable for Jetson
nvbuf-memory-type=0

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=2
uri=file:/opt/nvidia/deepstream/deepstream-6.0/samples/streams/sample_1080p_h264.mp4
#uri=file:/home/nvidia/Documents/5-Materials/Videos/0825.avi
num-sources=1
gpu-id=0
# (0): memtype_device   - Memory type Device
# (1): memtype_pinned   - Memory type Host Pinned
# (2): memtype_unified  - Memory type Unified
cudadec-memtype=0

[sink0]
enable=1
#Type - 1=FakeSink 2=EglSink 3=File
type=2
sync=0
source-id=0
gpu-id=0
nvbuf-memory-type=0
#1=mp4 2=mkv
container=1
#1=h264 2=h265
codec=1
output-file=yolov4.mp4

[osd]
enable=1
gpu-id=0
border-width=1
text-size=12
text-color=1;1;1;1;
text-bg-color=0.3;0.3;0.3;1
font=Serif
show-clock=0
clock-x-offset=800
clock-y-offset=820
clock-text-size=12
clock-color=1;0;0;0
nvbuf-memory-type=0

[streammux]
gpu-id=0
##Boolean property to inform muxer that sources are live
live-source=0
batch-size=4
##time out in usec, to wait after the first buffer is available
##to push the batch even if the complete batch is not formed
batched-push-timeout=40000
## Set muxer output width and height
width=1280
height=720
##Enable to maintain aspect ratio wrt source, and allow black borders, works
##along with width, height properties
enable-padding=0
nvbuf-memory-type=0

# config-file property is mandatory for any gie section.
# Other properties are optional and if set will override the properties set in
# the infer config file.
[primary-gie]
enable=1
gpu-id=0
model-engine-file=yolov5s.engine
labelfile-path=labels.txt
#batch-size=1
#Required by the app for OSD, not a plugin property
bbox-border-color0=1;0;0;1
bbox-border-color1=0;1;1;1
bbox-border-color2=0;0;1;1
bbox-border-color3=0;1;0;1
interval=0
gie-unique-id=1
nvbuf-memory-type=0
config-file=config_infer_primary_yoloV5.txt

[tracker]
enable=0
tracker-width=512
tracker-height=320
ll-lib-file=/opt/nvidia/deepstream/deepstream-5.0/lib/libnvds_mot_klt.so

[tests]
file-loop=0

camera

USB camera

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI
type=1
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1
camera-v4l2-dev-node=0

CSI camera

file

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP 5=CSI
type=5
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1
camera-csi-sensor-id=0

videofile

Four identical files, using MultiURI:

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP
type=3
uri=file://../../streams/sample_1080p_h264.mp4
num-sources=4
#drop-frame-interval=2
gpu-id=0
# (0): memtype_device   - Memory type Device
# (1): memtype_pinned   - Memory type Host Pinned
# (2): memtype_unified  - Memory type Unified
cudadec-memtype=0

media stream

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP
type=4
uri=rtsp://admin:admin123@192.168.1.106:554/cam/realmonitor?channel=1&amp;subtype=0
num-sources=1
#drop-frame-interval=2
gpu-id=0
# (0): memtype_device   - Memory type Device
# (1): memtype_pinned   - Memory type Host Pinned
# (2): memtype_unified  - Memory type Unified
cudadec-memtype=0

Multiple USB streams

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP
type=3
uri=file://../../streams/sample_1080p_h264.mp4
num-sources=4
#drop-frame-interval=2
gpu-id=0
# (0): memtype_device   - Memory type Device
# (1): memtype_pinned   - Memory type Host Pinned
# (2): memtype_unified  - Memory type Unified
cudadec-memtype=0

Multiple CSI streams

[source0]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP 5=CSI
type=5
camera-csi-sensor-id=0
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1

[source1]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP 5=CSI
type=5
camera-csi-sensor-id=1
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1

[source2]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP 5=CSI
type=5
camera-csi-sensor-id=2
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1

[source3]
enable=1
#Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP 5=CSI
type=5
camera-csi-sensor-id=3
camera-width=1280
camera-height=720
camera-fps-n=30
camera-fps-d=1

Video processing

Object detection

# config-file property is mandatory for any gie section.
# Other properties are optional and if set will override the properties set in
# the infer config file.
[primary-gie]
enable=1
model-engine-file=../../models/Primary_Detector/resnet10.caffemodel_b30_int8.engine
#Required to display the PGIE labels, should be added even when using config-file
#property
batch-size=4
#Required by the app for OSD, not a plugin property
bbox-border-color0=1;0;0;1
bbox-border-color1=0;1;1;1
bbox-border-color2=0;0;1;1
bbox-border-color3=0;1;0;1
interval=0
#Required by the app for SGIE, when used along with config-file property
gie-unique-id=1
config-file=config_infer_primary.txt

Object tracking

[tracker]
enable=1
tracker-width=640
tracker-height=368
#tracker-width=480
#tracker-height=272

#ll-lib-file=/opt/nvidia/deepstream/deepstream-4.0/lib/libnvds_mot_iou.so
#ll-lib-file=/opt/nvidia/deepstream/deepstream-4.0/lib/libnvds_nvdcf.so
ll-lib-file=/opt/nvidia/deepstream/deepstream-4.0/lib/libnvds_mot_klt.so
#ll-config-file required for DCF/IOU only
#ll-config-file=tracker_config.yml
#ll-config-file=iou_config.txt
gpu-id=0
#enable-batch-process applicable to DCF only
enable-batch-process=1

Secondary classification after detection

[secondary-gie0]
enable=1
model-engine-file=../../models/Secondary_VehicleTypes/resnet18.caffemodel_b16_int8.engine
gpu-id=0
batch-size=16
gie-unique-id=4
operate-on-gie-id=1
operate-on-class-ids=0;
config-file=config_infer_secondary_vehicletypes.txt

[secondary-gie1]
enable=1
model-engine-file=../../models/Secondary_CarColor/resnet18.caffemodel_b16_int8.engine
batch-size=16
gpu-id=0
gie-unique-id=5
operate-on-gie-id=1
operate-on-class-ids=0;
config-file=config_infer_secondary_carcolor.txt

[secondary-gie2]
enable=1
model-engine-file=../../models/Secondary_CarMake/resnet18.caffemodel_b16_int8.engine
batch-size=16
gpu-id=0
gie-unique-id=6
operate-on-gie-id=1
operate-on-class-ids=0;
config-file=config_infer_secondary_carmake.txt

Video output

Tiling multiple streams

Single stream:

[tiled-display]
enable=1
rows=1
columns=1
width=1280
height=720

Multiple streams:

[tiled-display]
enable=1
rows=4
columns=2
width=1280
height=720
gpu-id=0
#(0): nvbuf-mem-default - Default memory allocated, specific to particular platform
#(1): nvbuf-mem-cuda-pinned - Allocate Pinned/Host cuda memory, applicable for Tesla
#(2): nvbuf-mem-cuda-device - Allocate Device cuda memory, applicable for Tesla
#(3): nvbuf-mem-cuda-unified - Allocate Unified cuda memory, applicable for Tesla
#(4): nvbuf-mem-surface-array - Allocate Surface Array memory, applicable for Jetson
nvbuf-memory-type=0

screen

[sink0]
enable=1
#Type - 1=FakeSink 2=EglSink 3=File 4=RTSPStreaming 5=Overlay
type=5
sync=0
display-id=0
offset-x=0
offset-y=0
width=0
height=0
overlay-id=1
source-id=0

videofile

[sink1]
enable=1
type=3
#1=mp4 2=mkv
container=1
#1=h264 2=h265 3=mpeg4
codec=1
sync=0
bitrate=2000000
output-file=out.mp4
source-id=0

media stream

[sink2]
enable=1
#Type - 1=FakeSink 2=EglSink 3=File 4=RTSPStreaming 5=Overlay
type=4
#1=h264 2=h265
codec=1
sync=0
bitrate=4000000
# set below properties in case of RTSPStreaming
rtsp-port=8554
udp-port=5400

Open the network stream in VLC: rtsp://192.168.0.118:8554/ds-test

osd

[osd]
enable=1
border-width=2
text-size=15
text-color=1;1;1;1;
text-bg-color=0.3;0.3;0.3;1
font=Serif
show-clock=0
clock-x-offset=800
clock-y-offset=820
clock-text-size=12
clock-color=1;0;0;0

streammux

[streammux]
##Boolean property to inform muxer that sources are live
live-source=1
## Set this based on the number of sources
batch-size=4
##time out in usec, to wait after the first buffer is available
##to push the batch even if the complete batch is not formed
batched-push-timeout=40000
## Set muxer output width and height
width=1280
height=720

Sample apps

DeepStream Sample App /sources/apps/sample_apps/deepstream-app

Note: This end-to-end sample demonstrates multi-camera streams through a 4-stage cascaded neural network (1 primary detector + 3 secondary classifiers) and shows tiled output.

DeepStream Test 1 /sources/apps/sample_apps/deepstream-t
DeepStream Test 2 /sources/apps/sample_apps/deepstream-test2

Note: A simple app built on top of test1, showing additional features like tracking and secondary classification metadata.

DeepStream Test 3 /sources/apps/sample_apps/deepstream-test3

Note: A simple app built on top of test1, showing multiple input sources and batching via nvstreammux.

DeepStream Test 4 /sources/apps/sample_apps/deepstream-test4

Note: Built on top of the Test1 sample. Demonstrates use of the “nvmsgconv” and “nvmsgbroker” plugins in IoT-connected pipelines. For test4, you must modify the Kafka broker connection string to connect successfully. You need to install the analytics server Docker before running test4. See the DeepStream analytics docs for more details.

FasterRCNN Object Detector /sources/objectDetector_FasterRCNN

Note: FasterRCNN object detector sample.

SSD Object Detector /sources/objectDetector_SSD

Note: SSD object detector sample.

Deploy your own model

Deploy your own YOLOv5 model on Jetson Nano (TensorRT acceleration) (CSDN)

What are You Looking for?

YOLOv5 TensorRT DeepStream Deployment on Jetson Nano Guide

Introduction

Environment

On the Host PC

On the Jetson

Set up the YOLOv5 environment

Clone TensorRTx

Install and test DeepStream (5.0)

Install DS-6.0

YOLOv5 detection

Use CSI or USB cameras in DeepStream

Plugin configuration

Video input

camera

videofile

media stream

Multiple USB streams

Multiple CSI streams

Video processing

Object detection

Object tracking

Secondary classification after detection

Video output

Tiling multiple streams

screen

videofile

media stream

osd

streammux

Sample apps

Deploy your own model

References

Deployment

DeepStream

Read Next

jetson nano environment configuration

HPM5321 PCAN DIY Guide: Build Your Own USB to CAN Bus Analyzer

How Small Businesses Can Use AI to Compete with Big Companies in 2026

How xMemory Solves AI Agent Context Bloat: A Complete Guide for Enterprise Architects