How AI Photogrammetry Is Changing 3D Forever
Vložit
- čas přidán 13. 05. 2024
- Thanks to Babbel for sponsoring today's video. Go to strms.net/babbel_bycloudai to get up to 60% off! #ad
Intro Clip by Arata Fukoe
[Twitter] Arata_Fukoe/statu...
Instant-NGP
[My Video] • NVIDIA Just Made AI Ph...
[Project Page] nvlabs.github.io/instant-ngp/
[Paper] nvlabs.github.io/instant-ngp/...
[Code] github.com/NVlabs/instant-ngp
LERF
[Paper] arxiv.org/abs/2303.09553
[Project Page] www.lerf.io/
[Code] github.com/kerrj/lerf
[NeRF Studio Page] docs.nerf.studio/en/latest/ne...
MixNeRF
[Paper] arxiv.org/abs/2302.08788
[Project Page] shawn615.github.io/mixnerf/
[Code] github.com/shawn615/MixNeRF
Zip-NeRF
[Paper] arxiv.org/abs/2304.06706
[Project Page] jonbarron.info/zipnerf/
F2-NeRF
[Paper] arxiv.org/abs/2303.15951
[Project Page] totoro97.github.io/projects/f...
[Code] github.com/totoro97/f2-nerf
Block-NeRF
[My Video] • NeRF With Style Transf...
[Paper] arxiv.org/abs/2202.05263
[Project Page] waymo.com/research/block-nerf/
BungeeNeRF
[My Video] • NeRF Is Gonna Be The F...
[Paper] arxiv.org/abs/2112.05504
[Project Page] city-super.github.io/citynerf/
[Code] github.com/city-super/BungeeNeRF
GridNeRF
[Paper] arxiv.org/abs/2303.14001
[Project Page] city-super.github.io/gridnerf/
[Code]
NSFF
[My Video] • 3D Video Stabilization...
[Paper] arxiv.org/abs/2011.13084
[Project Page] www.cs.cornell.edu/~zl548/NSFF/
[Code] github.com/zhengqili/Neural-S...
ReLight My NeRF
[Paper] arxiv.org/abs/2304.10448
[Project Page] eyecan-ai.github.io/rene/
MSNeRF
[Paper] arxiv.org/abs/2305.04268
[Project Page] zx-yin.github.io/msnerf/
[Code] github.com/ZX-Yin/ms-nerf
Nerfbusters
[Paper] arxiv.org/abs/2304.10532
[Project Page] ethanweber.me/nerfbusters/
[Code] github.com/ethanweber/nerfbus...
Instruct 3D-to-3D
[Paper] arxiv.org/abs/2303.15780
[Project Page] sony.github.io/Instruct3Dto3D...
InstructPix2Pix
[Paper] arxiv.org/abs/2211.09800
[Project Page] www.timothybrooks.com/instruc...
[Code] github.com/timothybrooks/inst...
3D Video Loops from Asynchronous Input
[Paper] arxiv.org/abs/2303.05312
[Project Page] limacv.github.io/VideoLoop3D_...
[Code] github.com/limacv/VideoLoop3D
DyNeRF
[Paper] arxiv.org/abs/2103.02597
[Project Page] neural-3d-video.github.io/
NeRFPlayer
[Paper] arxiv.org/pdf/2210.15947.pdf
[Project Page] lsongx.github.io/projects/ner...
[NeRF Studio] Available
HyperReel
[Paper] arxiv.org/abs/2301.02238
[Project Page] hyperreel.github.io/#teaser-v...
[Code] github.com/facebookresearch/h...
Patch-based 3D Natural Scene Generation from a Single Example
[Paper] arxiv.org/abs/2304.12670
[Project Page] weiyuli.xyz/Sin3DGen/
[Code] github.com/wyysf-98/Sin3DGen
DreamFusion
[Paper] arxiv.org/abs/2209.14988
[Project Page] dreamfusion3d.github.io/
ProlificDreamer
[Paper] arxiv.org/abs/2305.16213
[Project Page] ml.cs.tsinghua.edu.cn/prolifi...
[Code] github.com/thu-ml/prolificdre...
HiFA
[Paper] arxiv.org/abs/2305.18766
[Project Page] hifa-team.github.io/HiFA-site/
[Code] github.com/HiFA-team/HiFA
Neuralangelo
[Paper] research.nvidia.com/labs/dir/...
[Project Page] research.nvidia.com/labs/dir/...
Luma Labs AI (Not Luma AI Labs oops)
[Web] lumalabs.ai/
[App Store] apps.apple.com/in/app/luma-ai...
This video is supported by the kind Patrons & CZcams Members:
🙏Andrew Lescelius, Alex J, Chris LeDoux, Alex Maurice, Tony Jimenez, Panther Modern, Jake Disco, Demilson Quintao, Shuhong Chen, Hongbo Men, happi nyuu nyaa, Carol Lo, Mose Sakashita, Miguel, Bandera, Gennaro Schiano, gunwoo, Ravid Freedman, Mert Seftali, Mrityunjay, Richárd Nagyfi, Timo Steiner, Henrik G Sundt, projectAnthony, Brigham Hall, Kyle Hudson, Kalila, Jef Come, Jvari Williams, Tien Tien
[Discord] / discord
[Twitter] / bycloudai
[Patreon] / bycloud
[Music 1] Skies - Slipstream
[Music 2] Celery Fields - Slipstream
[Profile & Banner Art] / pygm7 - Věda a technologie
![NVIDIA Just Made AI Photogrammetry 1,000x Faster [Instant-NGP]](http://i.ytimg.com/vi/fvXOjV7EHbk/mqdefault.jpg)
![NVIDIA Just Made AI Photogrammetry 1,000x Faster [Instant-NGP]](/img/tr.png)
![How Did Open Source Catch Up To OpenAI? [Mixtral-8x7B]](http://i.ytimg.com/vi/PYZIOMvkUF8/mqdefault.jpg)
![These TikTokers Are AI Generated [SD Multi-Frame Rendering]](/img/n.gif)
Go to strms.net/babbel_bycloudai to get up to 60% off! #ad
and I am also thinking to do a 100K Q&A, feel free to ask me questions here or on my Discord's special Q&A channel! discord.gg/NhJZGtH
congrats on 100k subs!
it feels like the 90s again.... everything moving so fast, a true inflection point in history. Then, the spread of the internet. Now, this stuff. Arbitrary content generation will be spontaneous and nearly free.... the world will be distinctly different before and after.
True that, things have skyrocketed to the point that no one can truly keep up. It's exciting but also concerning, like how social media blew up and changed society, both for better and worse. Are we Icarus flying into the sun, or if it's like opening Pandoras box and unleashing endless opportunity.
so true
It's free for now but not for long
@@mmmuck open source is always free
@@tuna1867 That's why companies like ChatGPT are trying to sell AI as a service instead of as a program. They are working to cull open source unfortunately
I cant believe it.
VR Movies are even closer than I expected.
Imagine a full blown movie that doesnt look "rendered" and you sit in the middle and can look around.
Mixing nerf with real time cgi might be the real deal.
I asked some guy to make a colab for nerf when the first paper came out, it was barley functional but when it worked, wow it worked. Now its easy as shit
This will be revolutionary to AR and home assistant robots when they become mainstream
_Good lawd,_ this is moving so quickly! I love how much of this is open-source, rather than being locked up in ivory towers.
Following your channel has been a real journey - thinking and writing about NNs has completely taken over my life over the past two years, and without your videos there are so many developments that I would probably have missed.
(Also: Oh boy, so 3d artists were just one or two years behind in terms of significant aspects of their work becoming automatable compared to 2d. I honestly thought there was gonna be a bit more of a grace period than that.)
your definitely not a 3d artist
Nice video. At this point I think I enjoy your content slightly more than 2 minute papers for the added memes 😅
Also he talks about the paper unlike two minute papers
And also slightly less annoying
Also he makes videos of papers that are not 6 months old already.
As a phd student, I really liked this video since it made me discover many papers that I hadn't heard of and it gives a very good overview of the current state of the art in the field. However, it's a bit depressing to see this crazy amount of competition. If it looks like a rush from an outsider's point of view, imagine what it's like from a researcher/phd student's point of view!
do you feel depressed you didn't invent some of this stuff yourself?
@@nicholash1278 not really since research is about finding new ways to do thing, so there is always something new to be tried. As mentioned in the video, some papers are published with just a few week intervals and yet improve on the state of the art by a wide margin. Having your hard work made obsolete just after you've published, or worse, before you publish, that's the depressing part. It hasn't happened to me yet but I'm sure that feeling is shared among all researchers and PhD students.
@@nicholash1278 PhD Students feel depressed when they realize what they are doing is so crap compared to state of the art, and may never be better than it. Pretty desparate if you choose the wrong subject and not talented enough.
I LOVE your videos! Thank you for the hard work!
This is a really terrific video. Thanks for making this.
You are awesome! thanks for all those great news!
This tech is advancing so fast these days 😅
This + Unreal 5. Oh boy indie devs are eating well
nah, they are going out of business because there will be too much competition bringing forward games that all look incredible. I think ai will be the death of indie games, because the market will become over-statured with games featuring this ai content. It's kind of depressing.
@@nicholash1278 Best possible outcome, as it won't be indie devs that die because of this, it'll be aibros who think just because their game looks good, that it is good who will.
@@QuincyIsCrispy AI bro is such a stupid term tbh
@@nicholash1278 Perfect fit for them then
Very good content man, keep it going
THIS IS AWESOME, I UNDERSTOOD SOME OF THE WORDS YOU SAID 🌟🌟🌟
I would be curious where the limits of this tech is, I've got Terapixel scale image sets of a town from a few different angles, would be interesting if that is enough to model most of the town in 1 shot, (Shot for panoramas, but data is still data)
With the 3D CLIP stuff it would be interesting to see if it has any effects in forensics where you could scan a house and search for common "bad items" and see if you missed anything or anything else could be inspected.
This tbis is so amazing 🔥🔥
Thanks for the video! What paper/video does the source video from 0:32 - 0:40 come from?
Great video!
wow this is a lot of progress... nice video by the way
Thanks for the videos! Bycloud
Generative 3d nerf that is trained on tonnes of images and real world 3d data is the holy grail. It will be used to improve all the detail in stuff like Google Earth. It will be so good that it could even be used in GTA 7 with the entire planet in high detail (streamed like MSFS2020 of course).
I don't understand why this AI tech is so underated compared to Chat GPT and even Dalle 2?😮💨
This thing is game changing!
@@kmouratidis Yeah, because people don't play video games and 3D is really scarce.
What a time to be alive!
Thanks for the videos, always interesting.
Which software are people using to create LERF images?
I would like to use it with motion control to create anime
My favorite nerf is RAW nerf. Idk if other nerf techniques can do this to but they have shown that it can change brightness, fov, focus and more.
basically this has the potential allow AI to become more sentient of their surroundings on a dynamic basis, being able to virtually map an entire space and identify any object scanned within it with high fidelity. It will also allow for AR to use this same data to create highly realistic and lifelike environments.
any ai that can map flat area to correct scale such as road section
what a day to be alive!
i want the camera of bladerunner... having a video, stop a frame, then rotate, zoom... play XD
Just a photograph would be awesome too... bladerunner (1st one)
Love your work byCloudAI
Can anyone answer this question for me can you use this to make into a 3d printable mesh model?
Awesome video ! BTW, it's Unreal Engine that Luma has a plugin currently for, not Unity :-)
I would love to see example of how to do this like a tutorial or something
boy this is a bit scary, I was very much excited when the first img generative tool came out. But now I feel like we are bulding hands, eyes, nose and ultimately a real functioning brain to AI
6:43 I think you meant depressed
This is FOOKIN EPIC!
❤❤❤ I love your videos. Thank you very much!
This is gonna be awesome for games. Holy crap
we are coming to the point that was described in 70s scifi, where a machine can do anything creative but you need operators which are educated, cultivated and have a big mental horizon to handle what the ai gives them. otherwise we will probably spiral downwards a slide of low effort and superifical stuff.
lol
I see this technology been used in movies, games,...
Like a real 3D without glasses
I hope they come up with a way to 3d print images from text to image prompts. That would be cool😊
which ai tubers do you watch?
i used to watch 2 min papers but the guy is slow and repeats his previous videos content like way too much
i just watch ai explained and u
u remind me of a combination of fireship + 2min papers
good job bruv keep it up
I mean I like the progress we make in research, but as a compsci student specializing in computer vision, I hate that everything is done with neural networks. Training takes long, you don't know what exactly what is doing, trying out changes can take ages if you have to retrain the model and lastly, I am just a fan of algorithmic processing of data, where I understand exactly how and why something works, but neural networks are still just a black box we still can't even get close to fully understanding.
And never will, it will always be hidden secret for us as predicted by Stas Lem in 1960s in Summa of Technologia (author of Solaris).
Not to mention the potential errors such as the "ai" cancer image thing where ot labeled any image with a ruler as cancerous.
Applied to this video, the part where it shows different cleaning agents, imagine it picking the wrong one and either destroying your house surfaces or something incredibly toxic.
What's worse, is that software is becoming more like neural networks - so large and complex that no one knows what's happening or how many microservices are being used. Architecture is designed around existing APIs and tools. Testing happens in production.
Software in the '80s used to be like a walled garden, where you could plan everything. And now it's a rainforest.
But our mentality hasn't updated as fast; there's still an expectation that you should understand _everything_ in your company's stack. But that's unrealistic and unnecessary - just learn enough to deliver what you need.
EDIT: I stole all of this from the talk "Observability for the 99% developers" by Jean Yang.
So which one is easiest to implement in Blender, and also free?
Now the A.I are coming to spoil our Where's Waldo books! Where will it end?!
Hey! Do you have your script for this video? Thanks
This is incredible
No easy Android app such as the one from Luma Labs AI?
Freeze,Rotate,Zoom,Focus,Enhance is becoming a reality 😳
but can 2 nerf objects collide with each other?
Your videos are great and very informative! It would be great if you spoke a bit slower at times.
Future videos would need to be made in hours if the research keeps accelerating
The thumbnail isn't doing this video justice. I skipped over it twice, but I'm glad the algorithm was insistent.
"Look what they have to do to mimic a fraction of our power" - human brain imagination.
Neuroangelo is blowing my mind! I have nothing else to say lol I am speechless over that one
as someone who wanted to become a 3d environment artist one day, it's a bit depressing to see.
i hope neuralangelo doesnt take days to render
Oh my goodness! The tech in RoboCop is becoming real in front of my own eyes!
Nice
all those solved captchas man...
take a picture in a haunted house then tell it to find the ghost
8:03 : Really looks like an iron throne from game of thrones, I see no difference
i need this
updated local nerf tutorial please
One day, the video will never end... 🤯
holy crap
imagine the procedural generation games could accomplish with more advanced versions of this.
Great video! More nerf!
8:03 Didn't watch game of thrones but that throne doesn't look iron to me
OUTDATED, it's 103k subs now
It's NeRF or nothin'!
Jokes aside. Very good idea. I hate tech videos that waste energy on being dynamic and funny.
*SINCE Apple* has a powerful “neural engine” (most powerful in any consumer SoC/CPU), I guess the company might use similar machine learning techniques to convert many 2D assets into stereo images for its new headset. The issue, though, the headset is too pricey for this to make an impact for any significant number of people.
Its not more powerful than the RTX on any nVidia. It might be more powerful in mobile SoCs. But I don't know about the new Hexagon from Qualcomm, not too far behind.
@@monad_tcp As I mentioned, we are talking about SoCs/CPUs, not GPUs. (That said, Apple’s solution is faster than what most GPUs have for this, too.)
@@tatianaes3354 how is their solution faster if you can't use it and the hardware is optimized for their specific model ?
Ive spent some time reverse engineering it and I call those claims bullshit.
@@monad_tcp What you can not use?
Apple has CoreML API and other ways to use Neural Engine. There is nothing faster in consumer world in SoCs/CPUs.
just think of the memes...
Imaging watching 3d movies using apple vision pro
YEAH P
Language models are the msg of machine learning
Everything ia revolutionary
Ok, but how are we on the waifu front?
Some day, someone, after maybe 🤔 7 years will come back to this video and will laugh 😂 at the quality of this ai🤖
This is dystopian
The neurovolution is here
This could be used to cheat on where's Waldo books
is nice
nice comments🙂
wow... just went I thought being a 3d artist would be safe from ai. Looks like I need to drop out of college (I am studying 3d) and pick a different field. Maybe construction worker.
This is ironic to have babbel as the sponsor as language learning is being attacked from both sides by LLM. One side, LLMs are excellent teachers so the need for a dedicated curriculum app is reduced and LLMs are excellent translators so that over time the need to learn another language will be reduced.
I dont get it but cool
I'm still just trying to identify boundaries grrrrr
I WILL rule the world, if everyone could just slow down a bit please.
How can you use this tech to scan 3D figures to Autocad and Solidworks ? 😳😳😳😳 @bycloud
Cyberpunk 2077 stuff
Thank you for (As always) great and informative videos :) happy to be your sub and watch each single video since "AI Converts Cartoon Characters To Real Life [Pixel2Style2Pixel]" :]
:)
Of course. Only a fool wouldn't add "Dino-Regularization."
So obvious...
Early
this technology in combination with the the new apple vision and we will soon be able to buy and sell braindances on the dark web :-D
Nerf and entirety of 3d is redudant. Nobody will (thankfully) need to do all that complex math in their head to position a cube near other cube, and write shaders for it (i mean - who is it even marketed to, harvard geniuses?).
In the future every single piece of media you`ll see will be text2video or text2image (or thought2stuff). 3d is a useless 80s fads that's gonna die down same way xbox Kinect did.
I could understand obe wird yoy were saying
No explainatuon of terms , just verbsl garbage