SparkFun sd-card not compatible with object_follow

Kits
1

Looks like the object_follow.ipnyb requires TensorRT=5.0. The sd-card that comes with the JetBot has TensorRT 5.1. I am re-imaging the card from the NVIDIA .zip to see if the problem goes away.

2

To be more specific, the ssd_mobilenet_v2_coco.engine on the SparkFun SD-Card is compiled for TensorRT 5.0. TensorRT 5.1 is installed on the SparkFun SD-Card. These two are incompatible and cause the object_follow notebook to fail. Is there a fix? Thank you.

3

this worked for me:

create a new SD card from scratch as outlined here:

https://github.com/NVIDIA-AI-IOT/jetbot … om-Scratch

with the following modifications:

this will give you TensorRT version 5.0.6.3 (instead of JetPack4.2.2’s 5.1) with Tensorflow and Jupyter etc already installed

  • download and unzip sparkfun’s modification jetbot folder

https://cdn.sparkfun.com/assets/learn_t … jetbot.zip

edit the setup.py file and replace the Adafruit requirements with SparkFun packages:

install_requires=[
‘sparkfun-qwiic-scmd’,
‘sparkfun-qwiic-micro-oled’,
],

a lot of the steps have already been included in the jetbot_image_v0p3p0.zip image, but make sure you run

sudo python3 setup.py install

and set-up the jetbot_stats service:

cd jetbot/utils
python3 create_stats_service.py
sudo mv jetbot_stats.service /etc/systemd/system/jetbot_stats.service
sudo systemctl enable jetbot_stats
sudo systemctl start jetbot_stats

i also did a

pip3 install smbus2

I included the TRT engine from the link provided here:

https://devtalk.nvidia.com/default/topi … und-error/

I’m using the 8265.NGWMG Intel WiFi Modules (802.11) instead of the Edimax EW-7611 ULB so I did a

sudo systemctl stop nv-l4t-usb-device-mode.service
sudo systemctl disable nv-l4t-usb-device-mode.service

to connect to wifi

on first boot of the image, i did a

sudo apt update && sudo apt dist-upgrade -y --autoremove

to get updates out of the way

hope this helps,

-Sean

4

Hey Sean,

thanks for solving my problem with object follow, but now i cant get my Edimax wireless working. lsusb shows the device is there and i’ve installed all the drivers but not having any luck with getting wireless enabled.

you mentioned that you are using 8265.NGWMG Intel WiFi Modules (802.11) instead of the Edimax EW-7611 ULB and stopped and disabled some services.

I tried reenabling and starting but still not having any luck.

Any help would be awesomely appreciated.

Sabri

5

nm…figured out

git clone https://github.com/lwfinger/rtl8723bu.git

$ cd rtl8723bu

$ export ARCH-arm64

$ make

$ sudo make install

$ sudo reboot

reference: https://edimax.freshdesk.com/support/so … spberry-pi

6

@[SeanLeary

No idea how you found one of the older images, but this is awesome! I am going to link this to the GitHub issue we filed to help other users.](https://forum.sparkfun.com/memberlist.php?mode=viewprofile&u=67089)

7

I’ll give this a try, but does NVIDIA provide documentation on how to train a model that’s compatible with the newer version TensorRT? Seems fragile if their pre-trained models only run on the exact HW/SW configuration as whatever the model was built on, unless there’s a clear process on how to train new models (granted this may take over a month on my $100 Jetson Nano).

8

From the GitHub issue I filed, they do have instructions for building a model… https://github.com/NVIDIA-AI-IOT/jetbot … -535186884

That being said, they released a new model to fix the TensorRT version issue (also linked in the GitHub issue).