Comparison of DeepStream SSD Object Detection Performance on NVIDIA Xavier NX and AGX Xavier

Comparison of DeepStream SSD Object Detection Performance on NVIDIA Xavier NX and AGX Xavier

Jetson AGX Xavier | Jetson Xavier NX

17 February 2021
WHAT YOU WILL LEARN?

1- Setting up the performance test

2- Copying the model and config files into the test folder 

3- Testing the performance in each device

ENVIRONMENT

Hardware 1: DSBOARD-NX2

Hardware 2:  Jetson AGX Xavier 

OS 1: JetPack 4.5 (L4T-32.5)

OS 2: JetPack 4.4 (L4T-32.4.3)



In this tutorial, we tested our NVIDIA Xavier NX and AGX Xavier's DeepStream performance in object detector sample. Our test sources are a video le (FPS:60, resolution:1920x1080, type:MP4) and 4 different cameras (FPS:25, resolution:1920x1080, type:RTSP).


NVIDIA Xavier NX Info:

Software: JetPack 4.5 (L4T-32.5)

Hardware: Jetson Xavier NX with DSBoard-NX2 Carrier Board DeepStream Version: 5.0.1

Power Mode: 15W 6CORE


NVIDIA AGX Xavier Info:

Software: JetPack 4.4 (L4T-32.4.3) Hardware: Jetson AGX Xavier Developer Kit DeepStream Version: 5.0.0

Power Mode: MAXN


Performance Test Setup

To begin with, let's set our test environment with maximum power mode: 

(for Xavier NX)


sudo nvpmodel -m 2

(For AGX Xavier)

sudo nvpmodel -m 0


Type these commands and continue other steps in both of them:


sudo jetson_clocks
cd /opt/nvidia/deepstream/deepstream/sources/objectDetector_SSD/
cd nvdsinfer_custom_impl_ssd
sed -i 's/CUDA_VER?=/CUDA_VER=10.2/g' Makefile
make
cd /home/nvidia/Downloads/
wget https://raw.githubusercontent.com/bower-media-samples/big-buck-bunny-1080p-60fps-30s/master/video.mp4


Then, copy sample_ssd_relu6.uff, ssd_coco_labels.txt and all configuration text files from the zip file which we gave from the top of the post into /opt/nvidia/deepstream/deepstream/sources/objectDetector_SSD/ folder.


Change source URL (for source0, source1, source2 and source3) in each deepstream_app_config_ssd_rtsp... file (3 files) to downloaded video file’s full path.




Xavier NX Test Results


The performance of each test command:

(12-13 FPS)


deepstream-app -c deepstream_app_config_ssd_fp32.txt


(12-13 FPS)


deepstream-app -c deepstream_app_config_ssd_rtsp_fp32.txt


(~36 FPS)


deepstream-app -c deepstream_app_config_ssd_fp16.txt


(~25 FPS - max FPS)


deepstream-app -c deepstream_app_config_ssd_rtsp_fp16.txt

cansend can0 123#1122334455667788

AGX Xavier Test Results


The performance of each test command:

(21-22 FPS)

deepstream-app -c deepstream_app_config_ssd_fp32.txt


(21-22 FPS)


deepstream-app -c deepstream_app_config_ssd_rtsp_fp32.txt


(55-56 FPS)


deepstream-app -c deepstream_app_config_ssd_fp16.txt


(25-26 FPS - max FPS)


deepstream-app -c deepstream_app_config_ssd_rtsp_fp16.txt


Comparison of Results

Xavier NX - Video Results (1080p, 60 FPS) FP32 -> 12-13 FPS * 4 (48-52 FPS)

FP16 -> ~36 FPS * 4 (~144 FPS)

Xavier NX - RTSP Results (1080p, 25 FPS) FP32 -> 12-13 FPS * 4 (48-52 FPS)

FP16 -> ~25 FPS (max) * 4 (~100 FPS)

AGX Xavier - Video Results (1080p, 60 FPS) FP32 -> 21-22 FPS * 4 (84-88 FPS)

FP16 -> 55-56 FPS * 4 (220-224 FPS)

AGX Xavier - RTSP Results (1080p, 25 FPS) FP32 -> 21-22 FPS * 4 (84-88 FPS)

FP16 -> 25-26 FPS (max) * 4 (100-104 FPS)



Thank you for reading our blog post. 


If you want to get updated on new blog posts, product launches and discounts, you can fill out the form and sign up for our newsletter. By signing up, you can reach various blog posts about AI, deep learning, machine vision, high-speed cameras, and industrial interfaces. 


You will receive a "free shipping" code for next purchase immediately.

English