Google Just Turned the RPi into a Supercomputer...
ฝัง
- เผยแพร่เมื่อ 8 เม.ย. 2023
- Coral.ai @raspberrypi = ???
Raspberry Pi 4 👉 amzn.to/3SBCRW0
Coral AI USB Accelerator 👉 amzn.to/3SBGrzM
Raspberry Pi Camera V3 Module 👉 amzn.to/3Szkgdn
This this video we run video object detection in realtime using the Coral AI USB accelerator stick.
coral.ai/docs/accelerator/get...
The Coral USB Accelerator adds a Coral Edge TPU to your Linux, Mac, or Windows computer so you can accelerate your machine learning models. This page is your guide to get started.
All you need to do is download the Edge TPU runtime and PyCoral library. Then we'll show you how to run a TensorFlow Lite model on the Edge TPU.
To learn more about the hardware, see the USB Accelerator datasheet.
www.raspberrypi.com/
From $35
You'll recognise the price along with the basic shape and size, so you can simply drop your new Raspberry Pi into your old projects for an upgrade; and as always, we've kept all our software backwards-compatible, so what you create on a Raspberry Pi 4 will work on any older models you own too. - วิทยาศาสตร์และเทคโนโลยี
More tech vids that will make you question everythingg 👉th-cam.com/video/HxhyH35Imkk/w-d-xo.html
"just", then demos a 4 year old product...
Man, it's a good video, but two years ago I've seen local high schoolers run face recognition algorithm on Raspberry Pi 3. They had a live demo and it seemed to work just well. No extra hardware required, just RP3 and camera.
@@erroneousbosch my reaction to the new bladee album
'Supercomputer' is a stretch. Supercomputer can do a million double precision million element matrix dot products in 5 ms, not just quantized int8 inference on pretrained models.
Lol, couldn't coral do face recognition/object with python tensor flow 2 years ago?
Yeah but probably this allows for offloading most of the ML computation off the processor which seems to be the point. Effectively what that will mean is that the pi will be able to more things than be dedicated pattern recognition based on opencv and tensorflow. Also keep in mind the clock speed of the pi is 1.5GHz so this should greatly improve the speed of computation.
More like 4 years ago, acctually.
Yeah, the man just woke up after 2019.
@@mamba0815a oh, yea...I just noticed that it was launched in March of 2019. So, maybe this video is reposted...
Clearly haven't seen it used with Frigate to be used for object and human detection
Google has had Coral TPUs out for a few years now. Too bad the USB version is really hard to find these days. Especially at MSRP.
We got one for free, an RPi camera, filters, lens, RPi 4 4G, 16GB SD Card and Sense HAT hat because of the "Astro Pi: Mission Space Lab" project where we make code in Python, then upload it to their website for it to then be passed over to the ISS to run that code.
@@joveaaron-real wow, that's so cool
The USB Coral TPU is like Raspberry Pi itself - if you're willing to pay 2-4x MSRP, they're available. If you expect to pay MSRP you're going to need luck catching a restock, or patience for a backorder to fill.
Could not wait to own the USB accelerator, I paid extra $100 on Amazon
@@joveaaron-real what is MSRP?
For those who just have a pi and a camera, you can do object tracking and recognition without coral. Quite a bit depends on your code and what you are trying to achieve.
tensor-lite you can have 5 fps on a webcam =)
I have been using the Coral TPU with Frigate for video object recognition (mostly persons and vehicles) with my security cameras for almost 3 years now. It is a nice little device.
What architecture do you use and how many fps do you get ?
How does it fare against a GPU?
Are you also doing indexing of your video?
@JeremyTesar I want to buy one to use with Frigate with 3 cameras. Do you think it can handle 3 cameras at 15fps?
These are fairly old kit. Yes, it does put the Raspi to its limits but the limits of a Raspi are very limited, if you can even get one. The challenge you will run into with these USB accelerators is the bandwidth over USB 3.0 which is about 500 MB/s after overhead. A modern laptop (I can attest to the Macbook Pro M2 Pro being quite good for learning AI) or a cloud AI IDE would probably be better for learning these days.
When Google released Coral AI, a Raspberry Pi cost $35. Now, with scalper pricing, it may be cheaper to buy a Cray. I guess the Pi will save you some electricity cost.😉
Lots of other Pis out there. Banana Pis etc. Also RISC V microcontrollers are starting to be used in that sort of form factor mini computer. The transition is moving fast.
Well as a 74 year old retired systems engineer recovering from 2 Strokes I couldn’t keep up with your get this and that. I only have an iPad 10.2 screen, however I did get that there’s a lot of new technology for the Raspberry Pi to evolve. But I recently bought an Elegoo Arduino Uno Super Starter kit to refresh my knowledge for Electronics and Programming. Thanks for sharing this video and information.
Thank you! I love my Raspberry Pi's. I've got a few of them. One of them is made into a tablet, but I found that the "Headless Raspberry Pi" much more useful without all the hardware.
I've done ML coding before in Matlab and Octave using my desktop computer. I'm so much more excited by these tiny computers that do so much for so little money.
Interesting video, though to be honest I missed this product when it launched. A few thoughts based on me recent experiences with AI:
Generally, a lot of AI you want to run will run on a modest, but modern CPU if needed: Support varies by model, but you can often configure your CPU cores as a Cuda device, and run these models slowly in software. Not necessarily useful for high speed applications in real time, but it can be great for processing data in a pinch. I will also note that I'm fairly confident a modern multicore CPU will outperform the Coral AI accelerator, as it's more of a small proof of concept device for getting an initial application working, than for high performance computing. I will also note that Coral AI only supports Tensorflow Lite, (at int8 quantization, and only for CNN models, I believe) which means that many models will require tinkering to function on one of these devices, which is certainly doable, but it's good to know that going in!
Also: Many popular models have great CPU support and plenty of optimizations. Llamacpp comes to mind, but other models such as GPT4 x alpaca, Pygmalion, and so on also have a variety of optimizations that help in running CPU based compute, such as 4 bit quantization, AVX2 support, and so on.
The other thing I'll note is that many models also support either a low VRAM parameter (Stable Diffusion run in Automatic1111's WebUI comes to mind), or attention slicing (Which essentially does the same thing. KoboldAI, a WebUI for large language models, does support this), and what they do is allow you to run part of a model in CPU + System RAM, as opposed to running it all on GPU + VRAM, allowing you to run large models with a surprising degree of competence, even with a modest computer, albeit often at a slower speed. You can also split attention between multiple GPUs with some Python-fu, though it's difficult to do so in a way that perserves performance without a high bandwidth interconnect like NVLink, so that's often limited to professional or extremely high end cards.
If I was to suggest a piece of affordable AI hardware for running popular models, you may want to do a video on using an Nvidia Tesla P40 in a small form factor build. It's affordable, supports popular frameworks, has a large VRAM pool, and uses less power than comparable modern cards (3090 and 4090 for instance), so it's a very reasonable entrypoint into machine learning, though it would require either a system with an iGPU, or a small dedicated GPU as the P40 does not natively support display out as I recall.
I saw another video where P40 👍 was mentioned as a decent starter AI engine with its large VRAM.
Coral would be fun to mess around with a bit if it were plentiful in supply as well as being cheap.
Kindest regards, neighbours and friends.
Your programing is next level and way above my pay grade.
seems like a great way to have your feed monitored. That was a nice thing about early RPI is seemed to be private.
The Coral USB Accelerator is almost 3 years old now.. What do you exactly mean with "Google just release its..????"
03:10 The detail parameters for the AI are OK for general recog,
but still lacking because according to it's plumage, the AI got the parrot ID wrong.
Admittedly it's a VERY common mistake, but once pointed out, few folks make the same gaf twice.
The 1st color bar on the bird's wing is GREEN, means that's a Red & Green Macaw (Ara chloropterus)
That same bar is a brilliant YELLOW on a Scarlet. ;)
You speak very clearly. Why the hard coded subtitles over the center of the video. Is very distracting (for me at least). Content was great. Cool tool.
Wow, no time wasted on an outro 😂
Now imagine building cluster of these
Nice. I just received mine today and cant wait to play with it ;)
I'm gonna push my raspberry pi to the absolute limit (by doing the computation on something else) and find out if it can handle one of the most challenging computations in all of computer science (citation needed)
Don’t bother. The answer is 42.
Been using this device on my Pi Zero for a few years now, good stuff
That's pretty cool. I've contemplated using AI on an ESP32 variant to spot sharks under water when I'm wingsurfing but it's just one project of many in my queue.
new ? , the Coral TPU as been out for years , but you are lucky getting one at the right price .
Used to do without any TPU, had to wait for 2 mins sometime 😅
Like the Intel Neural compute stick, this is too expensive for the hobbyist. I would prefer a PCIe version with multiple chips, although for the Pi you would need a compute module IO board which has a Gen2x1 port.
beautiful video!
Hello, witch program are you using for selecting on video, recod screen? Thanks
Hi, can you please tell me what case you use for Raspberry withthe camera module?
The corals been out for a long long time. Long before this video was released. It even had its first shortage before this was released. It’s typically used for small robotics but then was picked up and sought after for guys running facial recognition with security cameras.
I noticed near the end that if finally detected the dumbel.
I've had a USB-C Coral TPU laying around on a shelf for years now... don't know what your on about!
Great...another thing for me to buy...thanks
You can also buy an OAK-D Lite Camera that does the same thing.
The video was ok but the text in the middle of the screen made this very painful to sit through. Subtitles belong at the bottom of the screen.
Holy! ORANGE! PRS! I gotta start doing vids so I can buy these. Cheapest PRS I see here in Canada is around $4000 last I looked. Oh, thanks for bringing me up to speed on Coral. Saw it when it 1st was an idea but forgot about it till now.
Interesting video but those closed captions...... I have never seen anything like that before 😂
This is so cool... awsome project! I have a RPi on my drone that runs openCV to detect ArUco markers, but I only get like 10 FPS, at the most. I would love to try this Coral out. Have you tried any other edge TPUs? I hear Arduino has some as well, it would be cool to get your take on them.
Bro that first terminal setting up your cam and the coral. Was that just your native mac terminal?? Or Linux terminal? Looked like the Mac
"sudo su -" == "sudo -i"
lol. That escalated quickly he said
I'm so glad you didn't put annoying subtitles right in-Oh wait, never mind.
the Tiktokafication of TH-cam...
Object classification are already possible on RaspberryPie for quite some time without one of those TPU!
Yolo4 the last time I did it.
I have been trying to get a Google Coral usb for 2 years now at MSRP, at the current price they are too much. For now I am doing the same thing using an nVidia Jetson Nano.
Esp32 can do object recognition and it has just a fraction of rpi's power
As far as I know, it can only do basic object recognition, and only at low framerates. The Esp32 simply lacks the computing power to run deep learning algorithms. Also, many of the Esp32 object recognition projects online apparently involve streaming video to a computer to do the processing.
I made a real time emotion detector just using the raspberry pi 4 8gb, it is very possible without any extra hardware besides a camera
So most importantly, can it do “hotdog/not a hotdog”?
Hi there, great video! I would have a suggestion, I think you shouldnt put the live captions to the center of the screen, its pretty distracting. Also, I at least dont like seeing spoken text word by word, that feels a bit weird.
I kept my view to the lower half of the screen below the words. Everything above was too distracting
There are cheap dedicated SoC for IP cameras that have built-in AI accelerator engine already, eg SSC30KQ with 0.5T and SSC338G with 1.0T, its no Coral but should be enough for real time recognition?
What camera and case are those?
"just released a new product"
Yeah, Ford just released a new form of transportation, called the automobile.
ohh what a miss of movie quote opportunity,
Coral TPU - I came here to do just two things: look cool and do tensor calculations, and I'm all out of cool.
Quick question why do you need this board to do that, I have used tensor and just the pi camera and got same results?
This does it in hardware on an ASIC.
Good video .. how do you end your videos just pull the power cord on the recording device?lol
Hi, thx for this video. One question, what exactly Raspberry OS Version you are using? Bullseye? 64/32 bit? Thx in advance
Umm, yeah you could already do that with Doods, Frigate, Blue Iris etc, without even having the need for additional hardware
How can I buy one such unit? What is the price? Will it work on MS Windows and x86 Linux? Do they ship worldwide or it is an engineering test sample available only to participants from certain countries?
This Would be Great if it wasn't made by google. I wonder if there is an open Source one?
Nice video. One question though, how does the Coral Edge TPU compare with Intel's Neural Compute Stick? The Neural Compute Stick is a bit more expensive but seems to be more popular.
Intels Stick is wack don't use it
@@granatapfel6661 Is the coral edge TPU good then?
@@hewiweng99 yes
it's crazy how far we've come in just the past 8 months since this video was made...
Good video man...Sick!
I want to know if this is capable of running Large Language models? Can I purchase several of them, and run the LLM on multiple corals?
The ending was not weird at all.
My question is, would this device be sufficient for running Large Language Models on the Raspberry Pi, such as Vicuna, LLaMA, HuggingFace, or GPT4All? I mean both training or fine tuning the models and running them as the backend to a chatbot.
NO way. Training GPT even version GPT-2 requires huge data center. I have runned GPT-2 on 32GB RAM 8 cores Xeon dedicated server and response is about 30 second (without GPU)
Hahhahhhaaaahaaaaa!
Was interested, but had to move on due to the incredibly annoying big text all the time. If I want CC I'll turn it on.
have you tried Mythic AI analogic chips?
AI is right, the average modern building for living looks like a prison
ML / AI people, how are what I assume are OPENCV algos going to compare NN models.
Do not add text in the middle of the video please
So could I use this on my windows laptop?
The Hailo-8 Destroys the coral, but you need a special board for the cm4
Why thr subtitles in the middle? This is not tik tok.
new, isn't it in market for at least 2 years?
That’s been around for years. I don’t know where you’ve been on the raspberry pi.
do you think this could be used in conjunction with the Rabbit R1?
that building is a prison. the occupants robbed a bank by signing for a mortgage and now they have to pay the bank back. with interest.
Have you been in a coma, I have been using this for years. Certainly not new technology
Where's the rest?
Link to purchase coral?
this has been out for years. I've seen rpis using this in 2021.
why does the video end abruptly? theres no direction or plan for this video. it was rushed and low effort. the heck?
Clickbaits are one hell of a drug.
Turn off the flash card words - or I'll turn off.
> supercomputer
> 8MB of RAM
How certain was the Coral AI that the opposite building was a Prison?
You are about 2 years late with this dude. This has been around for a while.
Excelent tool for search alien and orbs craft
Bro, try training these same cnn models using coral. I would really like to see that. Cuz back in 2019 i used to do both, not only but both, 2D & 3D(msft Kinect - point cloud) realtime image(i.e. video) recognition on rpi-3b+ using "pretrained" models (exactly similar to how you have shown in this vid).
Does it allow for like running LLMs in raspi like Alpaca and Dolly 2.0? That would be an ibtresting experiment
i think it can't
LLMs are models with billions of parameters, and it will be difficult on this device.
@@kartikpodugu What about on Nvidia Nano Jetson?
@@konstantinrebrov675 The latest jetson orin should be able to do inference. first announced jetson nano might not be able to do inference also.
Llm can run on raspi with quantization for like 1 word every 10s, don't know woth coral
@@kartikpodugu Nope, just got an Orin, not seeing the 20 TOPS they say
I know I ordered mine more than a year ago and still haven't received it. Maybe his just got delivered...
I'v been trying to get one of these for almost 2 years - they're about as available as actual RPis !!! 🤑
Oh. No. very basic computing not supercomputer.
prision huh... might be right
the captions in the middle of the screen are pretty annoying
Subbed. I hope this thing can work with Raspberry Pi Zero because the 250-gram weight restriction in Europe and FAA will only allow explosive growth of a module for drone/robot products under 10 grams, pi itself and the module combined to 20 grams. The drone/robot mechanics will be 230 grams in a foreseeable future. But, I am pretty sure there are engineers somewhere California/China working on that 20 gram goal right now. It's nice that you can go get AI into your living room, but a drone/robot will go into your living room/balcony itself to do things. That will be true pervasion of AI.
Old stuff promoted new yeah yeah
Will it run GPT-J?
It's probably just an ESP32 MCU
When can I buy a quantum computer on amazon?
2045
What if it's plugged in to a old PC?
can it run GPT4All ??
Depthai oak works great with raspberry pi
good project, but this is the default source code, I think you can do your own project, maybe you can detect fire or missile attacks, and crash airplanes on your window :) sleep activity, street traffic ... I don't have money to buy it. Thanks for share.
Exactly! It's a $100+ Coral USB dongle with matrix computation hardware on it.