BORIStheBLADE
Getting comfortable
I wanted to see if users of the Jetson Nano recommend this device right now? Has its Deepstack software matured?
Nvidia Jetson Nano with Deepstack
- Thread starter BORIStheBLADE
- Start date
BORIStheBLADE
Getting comfortable
Anyone?!
t_andersen
Young grasshopper
I recently bought a Jetson Nano and installed it to work in parallel with the Win10 computer running BI5. I have not upgraded to BI6 because I read somewhere, that I would then need to make 24/7 recording at low resolution, and I am not happy about doing so. Maybe I misunderstood something.
The Jetson Nano works great, and for me it is much better than running Deepstack on docker on the Win10 computer. The inference time for an image in Jetson Nano is about 0.25 sec, which is somewhat better than on the Win10 computer. At a first glance such a computation time on the BI computer does not seem to be a big problem, but actually it was for me. If a spider web is swinging in front of a camera, you get triggers several times per second and then the Win10 computer becomes red-hot due to all of the Deepstack work, and any true triggers may be delayed by the myriad of images that must be analyzed. So the Jetson Nano really speeds things up, when there are many false triggers.
I am running the Jetson Nano "Head-less" at my remote site and access it from home using VPN to the remote router. I could not get Remote Desktop to work with Jetson Nano for reasons that I still do not understand, but the free "No Machine" works great, when I log into the Jetson Nano from a windows computer at home. I installed Docker with Deepstack for Jetson Nano and it worked right out of the box. I have written a Python program for Jetson Nano (using Pyzo IDE) that reads triggered low-res images from BI and looks after cars, persons, etc. It works well. The advantage of using your own Python program, is that you can tailor it to your needs. I have written a Python function that ignores cars that are parked, so I don't get false car triggers, when a spyder keeps giving false BI-triggers. It works well and fast. I have included a call to Pushover, so I get alarms on my Android phone with a certain cool-down time.
The biggest problem I have had is irritating and trivial. On Jetson Nano I need to mount a folder located on the Win10 computer, so the two computers can talk together. It should be easy with a "mount" statement in a Jetson terminal but I get all kinds of error messages, and I have spent several nights trying to solve the problem. I ended up with a work-around using samba on the file manager on the Jetson.
I have plans to set up a watch-dog in the operating system of the Jetson but not yet done it. Since this is at a remote site, I would also like at some time to get a remote reset feature for the Jetson.
Any feedback would be welcome, in particular to the issue of mounting a Win10 folder on the Jetson.
The Jetson Nano works great, and for me it is much better than running Deepstack on docker on the Win10 computer. The inference time for an image in Jetson Nano is about 0.25 sec, which is somewhat better than on the Win10 computer. At a first glance such a computation time on the BI computer does not seem to be a big problem, but actually it was for me. If a spider web is swinging in front of a camera, you get triggers several times per second and then the Win10 computer becomes red-hot due to all of the Deepstack work, and any true triggers may be delayed by the myriad of images that must be analyzed. So the Jetson Nano really speeds things up, when there are many false triggers.
I am running the Jetson Nano "Head-less" at my remote site and access it from home using VPN to the remote router. I could not get Remote Desktop to work with Jetson Nano for reasons that I still do not understand, but the free "No Machine" works great, when I log into the Jetson Nano from a windows computer at home. I installed Docker with Deepstack for Jetson Nano and it worked right out of the box. I have written a Python program for Jetson Nano (using Pyzo IDE) that reads triggered low-res images from BI and looks after cars, persons, etc. It works well. The advantage of using your own Python program, is that you can tailor it to your needs. I have written a Python function that ignores cars that are parked, so I don't get false car triggers, when a spyder keeps giving false BI-triggers. It works well and fast. I have included a call to Pushover, so I get alarms on my Android phone with a certain cool-down time.
The biggest problem I have had is irritating and trivial. On Jetson Nano I need to mount a folder located on the Win10 computer, so the two computers can talk together. It should be easy with a "mount" statement in a Jetson terminal but I get all kinds of error messages, and I have spent several nights trying to solve the problem. I ended up with a work-around using samba on the file manager on the Jetson.
I have plans to set up a watch-dog in the operating system of the Jetson but not yet done it. Since this is at a remote site, I would also like at some time to get a remote reset feature for the Jetson.
Any feedback would be welcome, in particular to the issue of mounting a Win10 folder on the Jetson.
There is no BI6 yet lol...now the latest versions of 5.4.5.x does allow one to do 24/7 recording of substream and switch to record to mainstream for the motion triggers, but you still have the option of recording 24/7 mainstream if that is your desire...
t_andersen
Young grasshopper
Sorry for my mistake about the BI version. I use ver 5.3.2.2, which then does not have the Deepstack interface. So is it true that with BI 5.4.5.x and Deepstack, I am forced to record the substream 24/7? I dislike that because I know that I will never look into it and it will just fill the harddrive and the NAS. I prefer high-resolution triggered clips only, but maybe I am overlooking something.