Benchmarks Targeted for Jetson TX2

Autonomous Machines Jetson & Embedded Systems Jetson TX2
,
1

Hello,

I’m studying about ‘the Autonomous landing of small UAV’ using jetson nano.
I built using SSD MobileNet v2 for it and done it.
But FPS is too low as 0.5~0.8 using SSD-MobileNet v2.
So I’ve changed the board to TX2 but it’s same.
Then, I’ve tried to using TensorRT, Benchmars etc.

I’ve just found the git-hub as below,
GitHub - NVIDIA-AI-IOT/jetson_benchmarks: Jetson Benchmark.

But I got the error msg as below, but I have no idea how to solve it.
Please help me.

============error=========================

kookmin-uav@ubuntu:~/jetson_benchmarks$ sudo python3 benchmark.py --all --csv_file_path /home/kookmin-uav/jetson_benchmarks/benchmark_csv/tx2-nano-benchmarks.csv \

                        --model_dir /home/kookmin-uav/jetson_benchmarks/models \
                        --jetson_devkit tx2 \
                        --gpu_freq 1122000000 --power_mode 3 --precision fp16

Please close all other applications and Press Enter to continue…
Setting Jetson tx2 in max performance mode
Traceback (most recent call last):
File “benchmark.py”, line 130, in
main()
File “benchmark.py”, line 28, in main
system_check.run_set_clocks_withDVFS()
File “/home/kookmin-uav/jetson_benchmarks/utils/utilities.py”, line 34, in run_set_clocks_withDVFS
self.set_clocks_withDVFS(frequency=self.gpu_freq, device=‘gpu’)
File “/home/kookmin-uav/jetson_benchmarks/utils/utilities.py”, line 74, in set_clocks_withDVFS
from_freq = self.read_internal_register(register=freq_register_, device=device)
File “/home/kookmin-uav/jetson_benchmarks/utils/utilities.py”, line 100, in read_internal_register
reg_read = open(register, “r”)
FileNotFoundError: [Errno 2] No such file or directory: ‘/sys/devices/gpu.0/devfreq/17000000.gp10bnafll_gpc0/rate’

3

Hi,

Please add --jetson_clocks to enable the clock with jetson_clocks instead.
For example:

$ sudo python3 benchmark.py --all --csv_file_path /home/nvidia/topic_232954/jetson_benchmarks/benchmark_csv/tx2-nano-benchmarks.csv --jetson_clocks

Thanks.

2 Likes
4

Thank you so much your help!
I’ve just make it run!

By the way, I have one more question.
I’d like to ask you this benchmark can use with pre-trained model by tf2?

Actually I trained SSD-MobileNet v2 by tf2 for UAV’s autonomous landing.
This model can detect helipad BUT FPS is appx. 0.5 with Jetson nano.
So I would love to make it higher but have no idea.
I’m trying to use Jetson-inference, too but I don’t know how to use my pre-trained model.
Not using benchmark is okay, If I can make FPS higher.
Please help me how to solve this problem.

Thank you so much!

5

Hi,

You can use trtexec for the benchmark.

Below is a tutorial that converts a TF2 model into TensorRT for your reference:

Thanks.

7

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