Nano 2gb impressions

First tho.. watch dronebot nano! 😁

So I kind of started down this around Christmas. I bought a raspberry Pi 4 after digging around a bit and seeing Google had some tensor flow os image examples just for a pi.

What I really wanted to get to was object detection unfortunately the only image they had required an integrated pi camera, and while cheap, I wanted to be able to use a USB camera so I set off to figure that out.

That's when the problems started.

After many hours and much pain it came down to installing CV2 in order to grab the image and video data. Maybe my mojo is just weak but after dependency and general failure after about a month on and off I gave up on that.

In that time I also swam around in the Google help and support and general ai ecosystem and I'm sorry but it is a train wreck. I could go on for a long time about this.

I love my raspberry pi, but it's just not made to do this, and my experience is the moment you step off the beaten path you lose.

To be honest I really didn't want to eat the personal Crow of this not working so buying the nano was kinda swallowing my pride but I am so glad I did.

I went small after being burned so I only got the 2GB model costing $59 at Amazon. I knew I would be crunching things so I got a heavy fan and the cheapest case I could find. at my door shipping tax and everything $100.

Waiting for my order I signed up for the Nvidia developer program and got immediately accepted.
they probably accept everyone.

I was immediately impressed with the support ecosystem. I registered for their forums and then found there's actually an app for it pretty cool.

They have a dedicated training section, the forums are on point, the only real downside is there is so much information. Nvidia is all up in the forums replying directly to consumers.. it is just a really good thing going on from my few brief encounters.

So my hardware finally arrives and I get everything set up.

There are two sets of videos I suggest everyone walk through and click along.. jetson ai fundamentals s1 and s3. s2 is jetbot and worth a watch but not core to the knowledge.

The documentation and resources for the videos is on GitHub and I ended up using my laptop to watch the videos and sometimes using SSH and sometimes using the desktop to control the nano.

The first set of videos uses Jupyter notebook and was my first exposure to that. I've been programming for a long long time so I guess I'm set in my ways and it really wasn't to my taste. (but it worked!)

Docker it's a big part of how they make this work easily and that was my first exposure to that as well. Make note of which jetpack you install you will need it later.

After getting my setup going and working through the first set of vids I had accomplished more than I had in a month trying to make it work on a raspberry Pi.

I have one video left on s3 and so far have easily used canned models with both images and video, created my own data sets and train them in a number of ways and I'm becoming more familiar and used to the entire process.

One big thing to note is even when the training videos aren't fast forwarding they will finish processing first sometimes by a lot.

I found it's not unusual to wait a good long time 5, 10 maybe up to 15 minutes for a process to finish.

Another thing that's really hard for me to get over is that there are many times when the computer appears locked up. The CPU process monitor is stopped everything is stopped sometimes there's even low memory warnings. I pushed the reset button many times before I figured out it's just this poor thing doing the best it can.. wait it out. So be patient. If you don't see an error let it go.

In the docker containers and in GitHub are many tools and resources which make this easy such as data capture tools and conversion tools and a step-by-step on how to do things in most places. The difference in the organization and support here is day and night.

So far I'm very very pleased.