There was a time when making a machine to identify objects in a camera was difficult, even without trying to do it in real time. But now, you can do it with a Jetson Nano board for under $60. How well does it work? Watch [Murtaza’s] video below and see what you think.
The first few minutes of the video piqued our interest, and good thing, too, because the 50 lines of code get a 50-plus minute video! It is worth watching, though, because there’s a lot of good information about how to apply this technique in your own projects.
Of course, there is a lot more than 50 lines of code doing the work. The Jetson code and OpenCV do a lot of heavy lifting. But it is a great thing to leverage powerful tools to make tough jobs easy.
We were impressed with the neural network’s ability to reason out images. There was an image of a person opening up a sofa into a bed and the system correctly identified it as both a sofa and a bed. Pretty impressive.
We’ve seen dedicated object recognition systems at this price point, but the Jetson has a lot of flexibility. We were excited to see the Jetson Nano and with 128 cores and a good bit of memory, we are waiting to see more really powerful projects built around it.
AI does not “reason”, AI infers.
Is that AI or AL (the OP)?
I’d like to build an autonomous robot cart for use at school. Looks like the Jetson is the best bang for the buck?
Do you reason using something other than inference rules?
What does it detect if you write “Ipod” on a paper and tape it to an apple? ;)
Not much will happen unless you put it in sight of the camera.
B^)
Any AI worth its neurons would detect an IPod Apple, rather than an Apple iPod. :)
Now we need to teach it to differentiate between Linux ( $var = “good”; ), and Mac OS ( $var = “bad”; ).
Speaking of which, does anyone know if Mac OS is still Linux on the inside? I used an original Macbook (2006) a while back and, once I was able to find the command line, all da good ol’ commands worked.
Way underneath it’s BSD based. So, yes, UNIX commands work well. Many shells are supported. I appreciate and use command line access a LOT.
Mac OSX is a hybrid of NetBSD, FreeBSD and OpenBSD. To be more specific, it is a proper UNIX (conforms to POSIX) and contains no Linux code at all. The Linux userland mostly conforms to POSIX which is what make it seem similar.
There has never been Linux in Mac OS, not even a tiny bit.
It detects Hot Dog and Not Hot Dog.
That was seriously one of the best Jin Yang moments of the whole series. And there were a lot of great ones!
Fantastic video. This guy is awesome!
If you want to be correct (read: pedantic) about it, yes, not a stitch of linux code exists in OSX.
When people think of linux, they think of the GNU project as closely intertwined with it, and they usually conflate the two. OSX ships with GNU bash and a few other GNU utilities, though as far as I’ve seen they’ve been slowly phasing them out as much as possible.
Here’s the open source components of Big Sur: https://opensource.apple.com/release/macos-112.html
As you can see they’re using pretty damn old versions of GNU utilities, and generally refuse to modify them, so that they don’t have to give out their source. Anyway, still, OSX lifts a few of its parts from the same sources (and licenses) as linux distros do, and therefore have the exact same programs as a linux distro does, despite UNIX in general, and specifically BSD distros with strict licenses do not.
Therefore, if you can accept that people typically confuse the term “linux” to mean distributions that include the kernel, and GNU utilities like bash and apache are typically a part of those distros, you can say that “parts of linux are in OSX.”
Q.E.D.
This video tells only half the story, and it’s the easy half at that. It does little to explain how to generate and train the model that is at the heart of this ML system. Generating and training the model would normally take place on a machine that is far more powerful than that little NVIDIA Jetson Nano dev board.
What we really need is a hot dog vs. not hotdog app.
The video is still there, but the available code and project has been removed and sits behind a paywall now…