@@martintomasek6097 These brands are competing against each other. If one does it everyone has to do it. You want to prove that your product has the best developers in the background and if you think it is possible, you want to be the first one who does it.
A very abridged version of an explanation of CNNs, a title that's giberish, and an ad for features that don't necessarily need a CNN and hypothetical features that the camera certainly doesn't have.
As a ML programmer I instincively rolled my eyes when a Motorcycle channel started talking about how AI works. F9 then proceeded to give one of the most succinct and understandable explanations of it that I've ever heard, and sprinkled it with fantastic jokes. Thoroughly impressed, have my sub. Even got me exited to check out this camera.
Glad to hear we kept your interest! AI Vision is about to become a massive part of motoring and it's not that complicated to understand. Also yes... your view just helped us fund a huuuuuge project ;) ~RF9
@@FortNine It's the kind of ad I like to see, mostly based on facts and very little on trying to change people's emotions so they feel good about buying something they don't need. Thanks for the thorough ad/review
@@alexruan5639and he enjoys the support of a professional team. Things become almost effortless, when you have a team of pros carrying you to the finish line. At least that's how it feels for me.
This is a master stroke in marketing. Attach the passion of motorcycles, to a lesson in technology, to an ad to a product that brings it all together. Bravo.
Ryan: This ace pro 2 now has a windguard, hard shell encasing the foam over the microphone…this illustrates the power of AI KH: The significance of the passage of time Both equally meaningfull.
Ryan could have a kitchen machine review channel and it would explain stuff in an interesting way. It's his way of going about subjects that makes his content valuable.
Ryan, i am a physics engineer and i really REALLY appreciate if you were to make more videos about neural networks, or statistics or any other math related subject because... You really have a way of explaining things that is hard to find.
Your explanation about AI makes so much sense, as a fellow physicist I don’t understand why it is hyped as much as it is. It’s not magic, it’s just math that we can now more affordably run thanks to advancements of the chip industry. Some less technical managers present it as the most groundbreaking invention of the millennium. Although they’ve said the same about NFT, blockchain, cryptocurrency, 3D printing, and so on
My feelings exactly. The first time I looked into CNNs and cost functions I remember having a big "huh, that's it?" moment. Not that it isn't useful technology! Just surprisingly uncomplicated. ~RF9
Honestly I'm more excited about the 3d printing than all of the other things combined. For one you can print your own panniers and make them to be modular I know people make headphones with thia tech; it's DMS project Omega
@@mehmetgurdal I wouldn't print panniers. I have a printer that probably could, but you run in to quite a few issues. 3D prints are almost never truly waterproof, so if you plan on subjecting the panniers to lots of weather, you just might end up with mold inside of the layers of the print. The plastic of choice for printing would be PETG imo, since Nylon and anything CF-reinforced is quite the hassle. You'd have to print them either in black and leave them that way or paint them, because most other colors are UV sensitive. If I was dead set on making my own hard panniers, I'd find a shop online that does lasercutting and glue/bond together aluminium panels that you bent in to the shapes you want. that also allows you to put secret conpartments in that no one will ever find.
I don't care that it was an ad. It'd happily watch it again if it were an hour long. Far more interesting than 99.9% of the other you tube stuff. Genius script and production. The effort put in to making this is very under appreciated.
As someone who has been doing computer vision and ML stuff for a while professionally, that was a very well done simplified introduction. (Yes, there are a few places where you simplified away some bits that are key, but really understanding those bits is not something you're going to get in a short description.) And a great description of edge detection and matrix convolution without needing to deeply understand matrix math!
I never thought i see a video about computer vision on YOUR channel, yet here we are. Amazing work, not too simplified and every time with a great demonstration. Loved it. Many regards from a biker that works in that field here in Germany 👋🙂
That is really all there is at a high level, Ryan clearly has a skill for communicating. Of course, it's _actually_ a lot more complicated going down, network and layer architecture, training data and bias, trying to avoid local minima, other gradient search options, all the incredible optimizations of that matrix math (cause yeah, they are a lot), but it's kinda wild how a ton of motorcyclists just learned how basic computer vision worked in a 13 minute camera ad. I may even use this in my classes! - a cs prof
AI summed up very nicely (paraphrasing at around 11:00) "It's not that they've made something artificial seem intelligent, what they've actually done is take our own intelligence and make it seem artificial". Mind=blown
Aside from anything to do with motorcycles or videos, I love that you are about the only person incorporating math and physics into explanations of things that are actually mathematical and matters of physics! Thank you so much-- Science Teacher.
Dude, I don’t even ride anymore for other reasons but I still watch all your videos because you don’t compromise and speak the truth regardless of trend, opinion or the lure of sponsors. Good for you!
This is a surprisingly good way to break down AI. It drives me insane how many people love or hate on AI without knowing a single thing about what it is, or how it works. It isn’t HAL9000, it isn’t sentient, it’s just a math function that looks for the way pixels on an image are organized, and does whatever the designers tell it to do when it sees specific patterns. Super helpful in a lot of ways, and I’d actually be okay if it were implemented in cars if it were able to match or beat Tesla FSD. That said, I don’t want it anywhere near motorcycles. Hallucinations are annoying in a car, but if my motorcycle randomly brakes, changes power delivery, or suspension stiffness when I’m not expecting it due to a hallucination, that could easily put me on the ground. The only implementation I would ever want might be a warning system that can monitor blind spots and warn me if someone is about to run me out of my lane, or if someone is about to pull out in front of me. I will always welcome tech that gives me more awareness, but I don’t want anything that removes or changes my control of the bike without my say so.
I've used those type of functions in a Big Rig Truck. I agree with you 100%. False braking, acceleration and endless buzzer's going off is very distracting and slightly dangerous. On a bike, it will get us killed.
I wish all ads were this good and watchable. With Ryan's smooth as room temperature butter delivery I am confident he could sell out a tractor trailer load of chest freezers in Tuktoyaktuk; in January... during an ice storm. The amount of shot planning going into this... congrats to the team!
Watch some comparisons videos. I don't think it's as simple the Insta360 being better at everything - it's mainly the night shooting and mic that is completely night and day difference and gopro is trash at but in ideal daytime conditions gopro is better at some things (or at least just as good). GoPro's main advantage is their durability is the best in the action cam market which, being an "action" cam, is a big box to tick and probably one of the main reasons they haven't gone broke yet
When a motorcycle vlogger explains AI better and more entertainingly than any AI maximalist.. Presentation skills are hard to beat ;) Congrats on another beautiful video.
It's amazing how concisely you explained all these key concepts. As a PhD student in robotics and AI, I think this contained almost all core ideas we had in our first image processing lecture back in the day. It's great to get this knowledge out there, so people stop falling for these marketing buzzwords. Absolutely priceless
wow, that's actually a pretty solid explanation of how convolutional neural networks work! very refreshing to see over "it does math magic", even if it was part of an ad read.
I’m studying engineering and this 13min camera ad just made me understand the working principle of deep learning better than anything I’ve seen before.
Well... I have now watched a 12 minute ad....and that's not a complaint. So so well done. I'll happily watch your next one and if I wasn't trying to stay within the DJI eco-system for my own channel I'd be strongly considering the Insta 360. I hope your contract with I360 is lucrative because ya'll have earned every penny!
Seems like many people are misinterpreting this video as just a 13 minute ad. While yeah, the video is sponsored by insta360, and yeah he does use one of their cameras, he does also talk mostly about how the AI features used pertain to motorcycling, and how you need to be wary not to be overcharged when it inevitably makes the large-scale market. He also mentions that the camera is decent for filmmakers, not as 'just' a dashcam or helmet cam. I do think the video could have used more examples of companies who have plans to integrate these kind of camera functions into bikes, if they haven't already done so, but it is generally a really informative video on how the vector maths enables the AI to function. In short, yes it's partly an ad, but it's also a cautionary and informative video on AI camera tech.
Am I the only one that felt the GoPro looked dramatically better for those night time shots? Better contrast, more realistic color, better dynamic range… it actually looked like you were riding at night. Still appreciate the tech in the Ace Pro 2 but from an artistic stance, that GoPro shot was just more appealing to my eye.
@@dull_Asian_dad Exactly: more info = more better for a computer vision application like self-driving or rider alert systems. For artsy stuff too, you can just colour it down if you want a moody shot. ~RF9
This is a masterclass in creating high quality content: difficult informations made easy to understand, humor, great visuals and quotes from top gear. Very well done, sir!
"Makes thousands of tiny illogical changes - we call this a political function." OMG! Both hilarious and devastatingly accurate! 😂👍 No wonder this is one of my favourite TH-cam channels.
Your narration is so smooth, incredible! It's a real pleasure to watch these videos because there's rarely any cuts and everything seems timed perfectly. Every word seems to be spoken at the exact right second to progress to picking up objects or looking at drawings. No mispronounciation, no cut - keep it up!
Vision input is a solved issue for about 30 years now. The real problem is output. How to communicate the results to the rider. It boils down to one thing - the Heads-Up Display. Extra input like a map or a speedometer on the helmet's visor isn't hard to do, but you need to take eyes off the road and focus on that extra information. What's really needed is a heads-up display capable of highlighting objects, which involves calculating the trajectory between the rider's pupil, coordinates on a visor and a real world object. And while it would be amazing to have a nearby car's front wheels outlined in red as soon as they turn into your lane, cheap HUDs are too imprecise to do that. Due to the short distance between eyes and the visor, a 1 mm error on the HUD mean as much as 50 cm error in the projected image. And motorcycle helmets aren't exactly known for staying in one place. The only HUD helmets known to work like that are the ones used by jet pilots, and there's a reason they cost €400k each. Obviously, it's military tech, so ~10x the actual price, but still - over a half of that is the HUD. They're made for a given pilot's head and eye distance to minimize the movement. It's still not enough - the pilot's face has to be constantly 3d scanned for relation between the visor and the wearer's eyes so that the tiny laser projectors could adjust in real time. Finally, it takes a minimum of 3 high speed cameras to support enough data for the system to calculate where exactly on a HUD the highlight should happen, and it has to be done about 180 times a second (otherwise the image is blurry). It's an entirely different thing to analysing a single video file during compression, which a single modern camera can do. HUD systems have to coordinate data from 3 different cameras while projecting an image on a visor precisely over a moving object in real time. The last time I checked (which was around 2020), it'd take an equivalent of 3 flagship mobile phones in hardware to get the image registering and computing alone, and it still wouldn't be enough (it'd then be capable of maybe 60 fps, nowadays probably closer to 120 fps). The laser projectors used in jet pilot helmets back then weren't commercially available at all - not even something similar. If that tech ever comes to motorcycles, it'll probably be first used during racing and priced at anything between $20k and $40k. If I were to guess, I'd say an alternative tech will emerge before that - something like a composite transparent screen actually used as a visor. I still wouldn't say it'll succeed because we aren't there yet battery-wise.
It is insane to me that you did a better job of explaining all of this AI stuff in a super engrossing motorcycle video than I have seen from any math or science youtuber.
This is a motorcycle channel who just taught me more about AI than anyone else, even while the whole video is basically an ad I actually enjoyed watching... and I despise ads.
As a software engineer I really struggle to find good ways to explain how AI actually works to laymen. Never did I expect to find it in a motorcycle video. Bravo
I've been fairly interested in AI for 40 years (yes, really -- I wrote a paper on it in high school -- graduated in 1985), and Ryan just taught me more about how it's being done these days than I've picked up anywhere else. Never a dull moment with Fortnine. Except when I'm waiting for the next video.
One of the best breakdowns of "AI" and how it works I have ran into so far but at the cost of watching a 13 min ad for a camera I will never buy. I hope the sponsor paid well, brand promotors of Ryan's skill level cost $$$$ if hired through an agency.
The editing and foresight with this video, once again is unparalleled and largely unnoticed by the viewing public. The very first scene is the video in summary: you can sit on a motorcycle, that bike can move you through your world, but most of us agree that this in not "riding a motorcycle!"
This is what I do for a living, I graduated with object recognition via edge detection. Never thought my hobby, motorcycling, would circle back into work
One of the best and simplest explanation of deep learning models I have found on the internet. Ryan's videos are always worth waiting and be excited for. Love your work !!
I absolutely didn't realize this was an Ace Pro 2 commercial until it was too late and I had already pressed "submit order" thanks again for taking my money away Ryan.
Ads are hated, so I want to hate this video. But I learned something, so my time hasn't been wasted. An educational video about AI, on a motorcycle youtube channel, just demonstrated perfect transactional value of advertisement. This is how we save the world.
you just explained conv-nural netowrks in most concise clear way I've seen, in a motorbike video. Gradient Decent in a fortnitne video.. really really good work. hats off
The videography and editing in this vid is spectacular. As always. And y'all find the coolest places to shoot! The night vision on that 360 is *worth* the ad promo. WOW. That's the closest to actual human night vision that I've ever seen from a consumer camera. It was nice to see your snarky grin again, Ryan, and seeing you having fun. Plus we get a bonus shot of another F9 team member! More of those, please!
Weird seeing Ryan talking about kind of my field of research in a video about motorcycles. But it was very well explained, not dumbed down to the point no one understands how it works, kudos to Ryan and the team!
As a developer that loves AI and motorcycles, this is right up my alley. Some Convolutional Neural Networks are being replaced by Vision Transformer (ViT), since they are more powerful and are context aware because they have global self-attention and use this self-attention mechanism to replace the need for convolutional kernels. All this requires is images that are labeled and it will figure it out what objects are based on the given context of the image. Very cool video, but people should know that this is not how LLMs, Diffusion Image models, or various segmentation models work. They use Transformer architecture that isn't talked about here.
Keep up the awesome work… you guys are moving so fast with everything you do… y’all blow my slow x gen mind with what you come up with to tell so much in such a short span of time and make it feel like it went by even quicker than the run time… with love and respect… keep on keeping on.
Latest marketing strategy for all companies… “Put “A.I.” in all the specs and charge 50% more.” It’s not true A.I., it’s machine learning. Common and not the same.
@@ethandavis7310 Why should it be defined only by people who claim to work in it? - The literal definitions of both words when separated: artificial: 1. made or produced by human beings rather than occurring naturally, especially as a copy of something natural. "her skin glowed in the artificial light" 2. (of a person or their behavior) insincere or affected. "an artificial smile" intelligence: the ability to acquire and apply knowledge and skills. "an eminent man of great intelligence" - It seems like people purposefully use both definitions of artificial when talking about A.I., yet the first definition does not fit at all if taken with the definition of intelligence; they (humans) have not produced something that has the actual ability to acquire knowledge. Therefore it is not actual A.I.. Artificial diamonds can still be said to be actual diamonds, even if nature didn't produce them. Then again can you even truly have knowledge if you don't also have actual consciousness? I've come to the conclusion that it is probably not even possible to create consciousness, and if I am right, then it seems it would also be impossible to create genuine intelligence, only a mimicry of such.
Simply amazing! Ryan's commercial breaks are more exciting and captivating than other top producers' main course. It couldn't possibly get any better... and yet the next video will most likely be yet another leap into new heights. Unbelievable!
10:47 just to add some context to this part, because I initially misunderstood what Ryan was saying and somebody else might, too: Training AI models can be crazy expensive. Just the training of GPT-4 was over $60 million. And as far as I know, that's not factoring in the hardware that was used for training, preparation of the training data, salaries, etc. And then you have marketing and all that good stuff. So yes, copy-pasting a finished AI model onto devices is almost free, but that's not what you are paying for. Just like good tyres, which might be like $20 in materials and manufacturing but then sell for more than $250 each. That's not them (just) being greedy, they are recovering the costs of RND. Paying a premium for AI is more than justified, and you shouldn't expect things with AI to have the same price as those without it. There are ofc products, which slap AI on the box and sell for triple the price. But that's paying for the label, not paying for the features. That's not what you should do. You should be paying for the Ace Pro 2, because of its great features, and not for the GoPro because it has GoPro written on it.
I have no interest in this camera, but I watched to whole thing because I know its the way to fund the channel and you guys deserve that. Cheers from Brazil!
I feel hoodwinked into watching a long advertisement...... yet..... it is Ryan and I learned stuff..... needing a shower after that.
feeling dirty but satisfied :'D
That’s almost every video on f9 and I’m here for it
@@ArmyOfThree1000 right on!!
this time felt creepy for me.
Just wait till next week, when I explain gyroscopic precession from inside your own shower. #SponsoredByManscaped. ~RF9
I have a degree in Data Science and do it professionally. This is probably one of the best mini-explainers for a deep learning pipeline I have seen.
so can you explain why is it better processed on board rather then in post production?
@@martintomasek6097
These brands are competing against each other. If one does it everyone has to do it.
You want to prove that your product has the best developers in the background and if you think it is possible, you want to be the first one who does it.
came here to say this. when an ML explanation starts with 'its all in the matrix' you know theyve done their research :p
Indeed. Only if my teacher would also have initially explained it in such a nice, easy way, I would have been spared a few hours of self study.
Computer Scientist here, and I couldn't agree more
50% ad, 50% really good intro to computer vision and ML models. What I took away is that I'll watch anything with FortNine production value
I live f9 videos but all you had to do to convince to not buy a thing is tell me it is “ai powered”.
I'm gonna need you to take about 20% off er Squirelly Ryan
Me too. I don't even know how to ride a bike.
A very abridged version of an explanation of CNNs, a title that's giberish, and an ad for features that don't necessarily need a CNN and hypothetical features that the camera certainly doesn't have.
Yeah, this is the same reason I gave up on this channel a while back. This is a great confirmation I made the right choice.
As a ML programmer I instincively rolled my eyes when a Motorcycle channel started talking about how AI works. F9 then proceeded to give one of the most succinct and understandable explanations of it that I've ever heard, and sprinkled it with fantastic jokes.
Thoroughly impressed, have my sub. Even got me exited to check out this camera.
There's millions of smart people, and millions of not so smart people.
But the most important people are those who can bridge the gap between them.
Was about to leave basically exactly the same comment
Yeah Ryan does these things on the reg, one of the reasons I’ll always watch an F9 video.
Well done ryan, you made me watch an 13min ad for a instacam and i was also intrigued watching it
Glad to hear we kept your interest! AI Vision is about to become a massive part of motoring and it's not that complicated to understand. Also yes... your view just helped us fund a huuuuuge project ;) ~RF9
@@FortNine It's the kind of ad I like to see, mostly based on facts and very little on trying to change people's emotions so they feel good about buying something they don't need. Thanks for the thorough ad/review
Was about to say the same. Nicely done insta360 ad that really made me want one.
Where's the discount code @@FortNine
Great video. I learned a lot I had no idea about. Thank you
I love how at 0:58 when ryan says "AI is not all it seems" the bike is metaphorically held down in the trunk
He also didn't drive it in the establishing shot
Only sat on it
That's what that meant
crazy that i studied AI for about 2 years and still fail to explain basic convolutional network in less than 3 min, yet you did it so effortlessly
He's researched, practiced, and rehearsed for this video. Presentation is another skill in itself! He's mastered it, so don't feel bad.
Study harder!
(just kidding bro)
@@alexruan5639and he enjoys the support of a professional team. Things become almost effortless, when you have a team of pros carrying you to the finish line. At least that's how it feels for me.
Came here to say this. I haven't seen such a CNN explanation even remotely as good as that one.
it's also a the-more-you-know problem. ask any good gardener about a supposedly simple thing as lawn care
This is a master stroke in marketing. Attach the passion of motorcycles, to a lesson in technology, to an ad to a product that brings it all together. Bravo.
He sold me on it!
this motorcycle channel explained computer vision and AI better than his tech peers 👍👍👍
Ryan would make a better science/engineering teacher than 90% of my professors.
The youtube Algorithm recommended me this video, your comment just made me aware of this fact LOL
Ryan: This ace pro 2 now has a windguard, hard shell encasing the foam over the microphone…this illustrates the power of AI
KH: The significance of the passage of time
Both equally meaningfull.
Ryan could have a kitchen machine review channel and it would explain stuff in an interesting way. It's his way of going about subjects that makes his content valuable.
Ryan, i am a physics engineer and i really REALLY appreciate if you were to make more videos about neural networks, or statistics or any other math related subject because...
You really have a way of explaining things that is hard to find.
@@pocotaligoswampfox4789 Like..... all of them? You only had bad teachers?
Agreed
loved how he went down the ramp to the skatepark, smashed the fancy V4 paniers on the concrete barrier and then maniacally chuckled :D
I was thinking how, if he had AI vision on his bike, it would have warned him that he was going to hit that wall.
It's also visibly dented by the end of the video nyahahah
@itoibo4208 Sometimes you know what you're going to hit, you just can't stop from hitting it :)
Further along in the video… He’s sitting on the seat of the bike, and you can easily see the size of the dent put in the saddle bag beside him
Your explanation about AI makes so much sense, as a fellow physicist I don’t understand why it is hyped as much as it is. It’s not magic, it’s just math that we can now more affordably run thanks to advancements of the chip industry.
Some less technical managers present it as the most groundbreaking invention of the millennium. Although they’ve said the same about NFT, blockchain, cryptocurrency, 3D printing, and so on
Its just statistics on crack with if clauses
Once you've explained it that way, it starts losing its magic
Hype because we can search for the best program to solve a problem :D and the fact it even is able to learn a representation AT ALL is cool
My feelings exactly. The first time I looked into CNNs and cost functions I remember having a big "huh, that's it?" moment. Not that it isn't useful technology! Just surprisingly uncomplicated. ~RF9
Honestly I'm more excited about the 3d printing than all of the other things combined.
For one you can print your own panniers and make them to be modular
I know people make headphones with thia tech; it's DMS project Omega
@@mehmetgurdal I wouldn't print panniers. I have a printer that probably could, but you run in to quite a few issues. 3D prints are almost never truly waterproof, so if you plan on subjecting the panniers to lots of weather, you just might end up with mold inside of the layers of the print. The plastic of choice for printing would be PETG imo, since Nylon and anything CF-reinforced is quite the hassle. You'd have to print them either in black and leave them that way or paint them, because most other colors are UV sensitive.
If I was dead set on making my own hard panniers, I'd find a shop online that does lasercutting and glue/bond together aluminium panels that you bent in to the shapes you want. that also allows you to put secret conpartments in that no one will ever find.
I don't care that it was an ad. It'd happily watch it again if it were an hour long. Far more interesting than 99.9% of the other you tube stuff. Genius script and production. The effort put in to making this is very under appreciated.
This is James Burke "Connections" level explanation of complex computational maths. Freakin' love it man. Nice work!
Thousands of tiny illogical changes = the political function. Pure gold.
I literally looked it up to see if it was real. 😂 His delivery was so honest.
@@josephcomingsis it real?
@@mehmetgurdal I guess he made a joke on polynomial function. You can express polynomials with matrices.
@@mehmetgurdal it is not. But it should be now.
I'm glad someone caught that!😂
Bro wtf how have you explained neural networks so simply and better than basically all other explainations I've heard and it's in a motorbike video???
That intro is gold!
it was, at first i thought he was riding a 3-wheel bike or something.
Its crazy to me that a motorcycle channel explained AI better than a lot of tech channels do. Great video.
As someone who has been doing computer vision and ML stuff for a while professionally, that was a very well done simplified introduction. (Yes, there are a few places where you simplified away some bits that are key, but really understanding those bits is not something you're going to get in a short description.) And a great description of edge detection and matrix convolution without needing to deeply understand matrix math!
Completely agree
I never thought i see a video about computer vision on YOUR channel, yet here we are. Amazing work, not too simplified and every time with a great demonstration. Loved it. Many regards from a biker that works in that field here in Germany 👋🙂
That is really all there is at a high level, Ryan clearly has a skill for communicating. Of course, it's _actually_ a lot more complicated going down, network and layer architecture, training data and bias, trying to avoid local minima, other gradient search options, all the incredible optimizations of that matrix math (cause yeah, they are a lot), but it's kinda wild how a ton of motorcyclists just learned how basic computer vision worked in a 13 minute camera ad. I may even use this in my classes! - a cs prof
Thank you, Ryan!
I’m a CV grad and I still do not think I learnt as much in class as I did in this video. Some people are just… different.
AI summed up very nicely (paraphrasing at around 11:00) "It's not that they've made something artificial seem intelligent, what they've actually done is take our own intelligence and make it seem artificial". Mind=blown
Don't you just hate catching a pannier on a concrete wall (5:58) leaving a nice visible dent (11:38).
Makes it feel more real (?)
Yes. ~RF9
That hysterical laugh when he stops after the hit
Yep felt that.
@@maxjah3524 We just won't mention it to Ducati. I had a bit of a chuckle.
Aside from anything to do with motorcycles or videos, I love that you are about the only person incorporating math and physics into explanations of things that are actually mathematical and matters of physics! Thank you so much--
Science Teacher.
"Hello I'm Ryan f9 " Absolute classic
As if we don’t know he’s he is yet 😂
@@JMKGarage yes, lol
😂
that was my favorite part. iconic
That line was crimped from a country singer, who started every concert with, “hello, I’m Johnny Cash.”
The way you explained machine learning… the matrix calculations, the slope of the cost function.. amazing
Thanks for a lecture on Machine Learning and thanks to the sponsor for bringing it to us.
Also, huge thanks to Edwin for the wonderful production!
Dude, I don’t even ride anymore for other reasons but I still watch all your videos because you don’t compromise and speak the truth regardless of trend, opinion or the lure of sponsors. Good for you!
As an engineer myself this is a masterpiece in so many levels. Thanks
im a data scientist and this is propably the most condensed explanation on computer vison and neural networks I ever heard! Nice work!
This is a surprisingly good way to break down AI. It drives me insane how many people love or hate on AI without knowing a single thing about what it is, or how it works.
It isn’t HAL9000, it isn’t sentient, it’s just a math function that looks for the way pixels on an image are organized, and does whatever the designers tell it to do when it sees specific patterns.
Super helpful in a lot of ways, and I’d actually be okay if it were implemented in cars if it were able to match or beat Tesla FSD. That said, I don’t want it anywhere near motorcycles. Hallucinations are annoying in a car, but if my motorcycle randomly brakes, changes power delivery, or suspension stiffness when I’m not expecting it due to a hallucination, that could easily put me on the ground.
The only implementation I would ever want might be a warning system that can monitor blind spots and warn me if someone is about to run me out of my lane, or if someone is about to pull out in front of me. I will always welcome tech that gives me more awareness, but I don’t want anything that removes or changes my control of the bike without my say so.
This is specifically machine learning
I've used those type of functions in a Big Rig Truck. I agree with you 100%. False braking, acceleration and endless buzzer's going off is very distracting and slightly dangerous. On a bike, it will get us killed.
As somebody working in IT since the mid-80s, I refer to today's "AI" as "glorified statistics programs". 😋
Sapient not sentient. Sentient (able to sense) is such a low bar it's not funny. Technically a fuel gage makes the bike sentient, or a dashcam.
@@rvaughan74 Hilarious point - never thought about that before. ~RF9
I wish all ads were this good and watchable. With Ryan's smooth as room temperature butter delivery I am confident he could sell out a tractor trailer load of chest freezers in Tuktoyaktuk; in January... during an ice storm. The amount of shot planning going into this... congrats to the team!
As sponsored content goes this was alright, in an intro to computer vision sort of way.
At first I was disappointed that this was just an advertisement. But then I learned a crap ton and found it really interesting. Great job!
So in conclusion: Insta360 Ace Pro 2 > Gopro Hero 13
Much, GoPro have lost the edge well and truly.
Just about any action camera is better than GoPro at this point.
Watch some comparisons videos. I don't think it's as simple the Insta360 being better at everything - it's mainly the night shooting and mic that is completely night and day difference and gopro is trash at but in ideal daytime conditions gopro is better at some things (or at least just as good). GoPro's main advantage is their durability is the best in the action cam market which, being an "action" cam, is a big box to tick and probably one of the main reasons they haven't gone broke yet
The daytime video quality looked horrible though
Except all the ace pro 2 day shots in this video looks trash with that artificial sharpening going on
When a motorcycle vlogger explains AI better and more entertainingly than any AI maximalist.. Presentation skills are hard to beat ;) Congrats on another beautiful video.
Only Ryan can make a ad that's more interesting and helpful than any other I have seen
so people watch ads to have fun, that'd explain many comments here
It's amazing how concisely you explained all these key concepts. As a PhD student in robotics and AI, I think this contained almost all core ideas we had in our first image processing lecture back in the day. It's great to get this knowledge out there, so people stop falling for these marketing buzzwords. Absolutely priceless
wow, that's actually a pretty solid explanation of how convolutional neural networks work! very refreshing to see over "it does math magic", even if it was part of an ad read.
As an engineer who works with AI, this is one of the best explanations of a convolutional network I've ever seen. Amazing work
Geez I hope insta360 paid well
They sure did, every big youtuber on this platform got enought to tell everyone to buy this camera that' basically the same of the Ace Pro 1
I’m studying engineering and this 13min camera ad just made me understand the working principle of deep learning better than anything I’ve seen before.
Just saw my car at 4:20. Amazing work as always
420 AAAAAAAAAAAY
Sick Golf. Sportwagon? ~RF9
@@FortNine Close but it’s an Alltrack. Same same but different
Well... I have now watched a 12 minute ad....and that's not a complaint. So so well done. I'll happily watch your next one and if I wasn't trying to stay within the DJI eco-system for my own channel I'd be strongly considering the Insta 360.
I hope your contract with I360 is lucrative because ya'll have earned every penny!
Cool intro going into a 13:26 minute commercial for a camera lol.
I didn't expect that motorcycle channel will explain AI better and with better accuracy than software channels. Impressive. Good job!
this is such a beautiful tour of Vancouver and North Van !
The science nerd in me liked that - thanks. To those complaining it is an advertisement, it is but only in part and it is done with RF9’s usual class.
You just willingly watched a 13 min ad, very well integrated.
hated it. very annoying
Interesting though isn't it, I've learnt something and don't have to buy anything.
@@kentalanlee thanks fixed
@@Dreddingbath wait until finding out about Wikipedia, you are in for a wild ride
This might be the only ad that I've watched voluntarily, and also bought something because of it. Thank you Ryan
0:38 I looked away for a split second and was really confused to se Ryan standing on a trailer
Never expected to watch - motorcycle video.. that is an action cam AD.... that has best explanation of AI i ever heard......
The diarrhea of self promotion! How I love this man!
Give your location scout a raise! You found some amazing spots for the story. Bravo.
Seems like many people are misinterpreting this video as just a 13 minute ad. While yeah, the video is sponsored by insta360, and yeah he does use one of their cameras, he does also talk mostly about how the AI features used pertain to motorcycling, and how you need to be wary not to be overcharged when it inevitably makes the large-scale market. He also mentions that the camera is decent for filmmakers, not as 'just' a dashcam or helmet cam. I do think the video could have used more examples of companies who have plans to integrate these kind of camera functions into bikes, if they haven't already done so, but it is generally a really informative video on how the vector maths enables the AI to function.
In short, yes it's partly an ad, but it's also a cautionary and informative video on AI camera tech.
the best and the simplest explanation for AI-CNN i have seen so far - great work !!!
Am I the only one that felt the GoPro looked dramatically better for those night time shots? Better contrast, more realistic color, better dynamic range… it actually looked like you were riding at night.
Still appreciate the tech in the Ace Pro 2 but from an artistic stance, that GoPro shot was just more appealing to my eye.
Looks like there are 3 of you. 😁
But which would you rather if you were trying to detect pedestrians stepping out in front of you?
You can get that look in post in editing by crushing the shadows down. You can't bring the detail back out from from the GoPro.
You can turn off the AI stuffs and the images should still be better than the GoPro due to its bigger sensor advantage.
@@dull_Asian_dad Exactly: more info = more better for a computer vision application like self-driving or rider alert systems. For artsy stuff too, you can just colour it down if you want a moody shot. ~RF9
This is a masterclass in creating high quality content: difficult informations made easy to understand, humor, great visuals and quotes from top gear. Very well done, sir!
"Makes thousands of tiny illogical changes - we call this a political function." OMG! Both hilarious and devastatingly accurate! 😂👍
No wonder this is one of my favourite TH-cam channels.
Your narration is so smooth, incredible! It's a real pleasure to watch these videos because there's rarely any cuts and everything seems timed perfectly. Every word seems to be spoken at the exact right second to progress to picking up objects or looking at drawings. No mispronounciation, no cut - keep it up!
Vision input is a solved issue for about 30 years now. The real problem is output. How to communicate the results to the rider.
It boils down to one thing - the Heads-Up Display. Extra input like a map or a speedometer on the helmet's visor isn't hard to do, but you need to take eyes off the road and focus on that extra information. What's really needed is a heads-up display capable of highlighting objects, which involves calculating the trajectory between the rider's pupil, coordinates on a visor and a real world object. And while it would be amazing to have a nearby car's front wheels outlined in red as soon as they turn into your lane, cheap HUDs are too imprecise to do that. Due to the short distance between eyes and the visor, a 1 mm error on the HUD mean as much as 50 cm error in the projected image. And motorcycle helmets aren't exactly known for staying in one place.
The only HUD helmets known to work like that are the ones used by jet pilots, and there's a reason they cost €400k each. Obviously, it's military tech, so ~10x the actual price, but still - over a half of that is the HUD. They're made for a given pilot's head and eye distance to minimize the movement. It's still not enough - the pilot's face has to be constantly 3d scanned for relation between the visor and the wearer's eyes so that the tiny laser projectors could adjust in real time. Finally, it takes a minimum of 3 high speed cameras to support enough data for the system to calculate where exactly on a HUD the highlight should happen, and it has to be done about 180 times a second (otherwise the image is blurry). It's an entirely different thing to analysing a single video file during compression, which a single modern camera can do. HUD systems have to coordinate data from 3 different cameras while projecting an image on a visor precisely over a moving object in real time.
The last time I checked (which was around 2020), it'd take an equivalent of 3 flagship mobile phones in hardware to get the image registering and computing alone, and it still wouldn't be enough (it'd then be capable of maybe 60 fps, nowadays probably closer to 120 fps). The laser projectors used in jet pilot helmets back then weren't commercially available at all - not even something similar. If that tech ever comes to motorcycles, it'll probably be first used during racing and priced at anything between $20k and $40k. If I were to guess, I'd say an alternative tech will emerge before that - something like a composite transparent screen actually used as a visor. I still wouldn't say it'll succeed because we aren't there yet battery-wise.
It is insane to me that you did a better job of explaining all of this AI stuff in a super engrossing motorcycle video than I have seen from any math or science youtuber.
This is a motorcycle channel who just taught me more about AI than anyone else, even while the whole video is basically an ad I actually enjoyed watching... and I despise ads.
As a software engineer I really struggle to find good ways to explain how AI actually works to laymen. Never did I expect to find it in a motorcycle video. Bravo
You should inform that video was sponsored at beginning, but you told us it anyway so you did better than most youtubers
I've been fairly interested in AI for 40 years (yes, really -- I wrote a paper on it in high school -- graduated in 1985), and Ryan just taught me more about how it's being done these days than I've picked up anywhere else. Never a dull moment with Fortnine. Except when I'm waiting for the next video.
One of the best breakdowns of "AI" and how it works I have ran into so far but at the cost of watching a 13 min ad for a camera I will never buy.
I hope the sponsor paid well, brand promotors of Ryan's skill level cost $$$$ if hired through an agency.
one of the best walkthroughs of classical computer vision and modern model approaches
The editing and foresight with this video, once again is unparalleled and largely unnoticed by the viewing public. The very first scene is the video in summary: you can sit on a motorcycle, that bike can move you through your world, but most of us agree that this in not "riding a motorcycle!"
I'm pleasantly surprised to hear such an intuitive yet accurate breakdown of machine vision from you guys. Wouldn't expect it to be a priority to you
This is what I do for a living, I graduated with object recognition via edge detection. Never thought my hobby, motorcycling, would circle back into work
Absolutely love the long one take video style. So interesting and the skill that goes into it is always mesmerising
10:30 honestly it diesn't even compare, the action cam shots are obvious in the video and I enjoy the normal camera shots a million times more.
Simplifide description of a complicated system, that is the best kind of education. Well done with this video.
Me learning my Machine learning course from FortNine. Now that's what I am talking about!! Thanks a lot man!!
One of the best and simplest explanation of deep learning models I have found on the internet. Ryan's videos are always worth waiting and be excited for. Love your work !!
I absolutely didn't realize this was an Ace Pro 2 commercial until it was too late and I had already pressed "submit order" thanks again for taking my money away Ryan.
The writing of this channel & overall production is UNREAL!!!
Ads are hated, so I want to hate this video. But I learned something, so my time hasn't been wasted. An educational video about AI, on a motorcycle youtube channel, just demonstrated perfect transactional value of advertisement. This is how we save the world.
you just explained conv-nural netowrks in most concise clear way I've seen, in a motorbike video. Gradient Decent in a fortnitne video.. really really good work. hats off
Glad you included the dent in the luggage ;-)
The videography and editing in this vid is spectacular. As always. And y'all find the coolest places to shoot!
The night vision on that 360 is *worth* the ad promo. WOW. That's the closest to actual human night vision that I've ever seen from a consumer camera.
It was nice to see your snarky grin again, Ryan, and seeing you having fun. Plus we get a bonus shot of another F9 team member! More of those, please!
Did he just bump the right box on that concrete wall at the skate park and giggle?!
Weird seeing Ryan talking about kind of my field of research in a video about motorcycles. But it was very well explained, not dumbed down to the point no one understands how it works, kudos to Ryan and the team!
I think this is the most well polished turd of a video that I've had the displeasure of watching half of. TLDR: buy their action cam
But only if you don't want to buy a separate wind muff for your mic, and only shoot during nighttime.
Bc that daytime footage just looked horrible.
@@quakslikeaduckYeah, that daytime footage was garbage, almost pastel like. Comparing it to a proper camera was distasteful at best
I learnt more about AI in this, than I did watching videos claiming to explain AI. Love it.
Am I going to learn backpropogation from RaynF9 today?
As a developer that loves AI and motorcycles, this is right up my alley. Some Convolutional Neural Networks are being replaced by Vision Transformer (ViT), since they are more powerful and are context aware because they have global self-attention and use this self-attention mechanism to replace the need for convolutional kernels. All this requires is images that are labeled and it will figure it out what objects are based on the given context of the image. Very cool video, but people should know that this is not how LLMs, Diffusion Image models, or various segmentation models work. They use Transformer architecture that isn't talked about here.
5:57 bag damage
Damage can be seen at 10:43
Emotional damage for Ducati :D
Yes its an advert. But also (probably) the best explanation of how AI works, on TH-cam. Awesome!
The first 3 to 4 minutes were a masterpiece!
Agreed - it really tailed off after that.
(joke)
Keep up the awesome work… you guys are moving so fast with everything you do… y’all blow my slow x gen mind with what you come up with to tell so much in such a short span of time and make it feel like it went by even quicker than the run time… with love and respect… keep on keeping on.
Latest marketing strategy for all companies… “Put “A.I.” in all the specs and charge 50% more.”
It’s not true A.I., it’s machine learning. Common and not the same.
ML is a subdomain of AI...this is common knowledge....
"True" AI isn't a topic of conversation among people who actually work in AI. This is AI.
@@ethandavis7310 Why should it be defined only by people who claim to work in it?
-
The literal definitions of both words when separated:
artificial:
1.
made or produced by human beings rather than occurring naturally, especially as a copy of something natural.
"her skin glowed in the artificial light"
2.
(of a person or their behavior) insincere or affected.
"an artificial smile"
intelligence:
the ability to acquire and apply knowledge and skills.
"an eminent man of great intelligence"
-
It seems like people purposefully use both definitions of artificial when talking about A.I., yet the first definition does not fit at all if taken with the definition of intelligence; they (humans) have not produced something that has the actual ability to acquire knowledge. Therefore it is not actual A.I.. Artificial diamonds can still be said to be actual diamonds, even if nature didn't produce them. Then again can you even truly have knowledge if you don't also have actual consciousness? I've come to the conclusion that it is probably not even possible to create consciousness, and if I am right, then it seems it would also be impossible to create genuine intelligence, only a mimicry of such.
Simply amazing! Ryan's commercial breaks are more exciting and captivating than other top producers' main course. It couldn't possibly get any better... and yet the next video will most likely be yet another leap into new heights. Unbelievable!
A 13 minute commercial that felt like a college lecture and motorcycle riding. I will stick with my low tech motorcycles and avoid high traffic areas.
this. ty. low cheap motorcycle = light, easy to turn & accelerate upto 120 kph
I don’t even ride motorcycles any more, and yet this channel, for me, is required viewing. Keep it up, as I’m all in. 🤙👊🙏🏻😎
10:47 just to add some context to this part, because I initially misunderstood what Ryan was saying and somebody else might, too:
Training AI models can be crazy expensive. Just the training of GPT-4 was over $60 million. And as far as I know, that's not factoring in the hardware that was used for training, preparation of the training data, salaries, etc. And then you have marketing and all that good stuff.
So yes, copy-pasting a finished AI model onto devices is almost free, but that's not what you are paying for.
Just like good tyres, which might be like $20 in materials and manufacturing but then sell for more than $250 each. That's not them (just) being greedy, they are recovering the costs of RND.
Paying a premium for AI is more than justified, and you shouldn't expect things with AI to have the same price as those without it. There are ofc products, which slap AI on the box and sell for triple the price. But that's paying for the label, not paying for the features. That's not what you should do. You should be paying for the Ace Pro 2, because of its great features, and not for the GoPro because it has GoPro written on it.
AI is usless and is wrong 99% of the time. It's NFT of 2024.
this .ty. good things cost money , not free...
Didn't expect to see a sobel kernel applied to a convolution function explanation in this channel. Love it.
That's just an Ad for a Camera...
You say "just" as if it's anything less than a brilliant motorcycle-themed primer to understanding machine learning and computer vision.
I have no interest in this camera, but I watched to whole thing because I know its the way to fund the channel and you guys deserve that.
Cheers from Brazil!