← Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Radiance Fields: The Next Leap in Visualization
00:38:26
Radiance fields are reshaping how we capture, simulate, and share the world. Michael Rubloff explains the future of 3D workflows and how to start today.
Loading summary
Transcript47 lines
- [00:05]
A
Welcome to Reshaping Workflows with dell Pro Max PCs and Nvidia, where innovation meets real world impact in high performance computing.
- [00:20]
B
Welcome back to another episode of Reshaping Workflows with Delpro Max and Nvidia RTX GPUs. I'm your host, Logan Lawler. So if you're back, that's wonderful because we've done quite a few of these episodes. So today we're kind of shifting gears a little bit. We have focused quite a bit, you know, in terms of industry workflows. We've had a few folks from M and E, we've had a few folks from just kind of a general AI like Comfy Media Entertainment Workflow. We've talked to a few folks with Data Science. But today we're kind of, I would say arguably shifting fields quite a bit. And Fields is interesting because we're here to talk about Radiance Fields. So my guest today is my ruble off. He is the owner manager of Radiance Fields.com but I don't want to steal all of this thunder. We had a chance to catch up at GTC and thought he'd be a perfect guest for the show. So Michael, take a second, introduce yourself, tell us all a little bit about your background and then we'll jump right into it.
- [01:16]
A
Sure, yeah. So my name is Michael Rublef and I am the founder and managing editor of Radiance Fields.com and basically what Radiance Fields are, there's a whole bunch of different types of radiance field representations. But what they're able to do is they're able to take a series of either 2D images or a video of something and you're able to reconstruct very, very lifelike 3D out of it. And so right now the two most popular ways to reconstruct a radiance field are nerfs, neural radiance fields, or there's now really 3D gauss and splatting for real time radiance field rendering, which essentially both of these things allow you to reconstruct the world into a very lifelike 3D capacity.
- [01:55]
B
Fantastic. I told you it's going to be different today. Like we're doing this is more, you know, think of manufacturing, think of aco, like architecture design anywhere where you have, you know, a physical space where you have kind of a 2D input and then how do you convert that to 3D interact? I'm moving my hands a lot, but 3D interchangeable, interactable thing. So let's start really simple for those. You kind of talked about Gaussian splattering, you talked about nerfs. But let's take a step back, you know, for. Let's say pretend you were talking to my mom and my mom is zero bit technical. What is a radiance field?
- [02:33]
A
So basically a radiance field is a way to show a given scene or moment in times actual. Like. Like radiance. That means like what. What does it look like from a given angle? And so these radiance fields are able to model with something is called view dependent effects. Which basically means, you know, if we take a look at something that's reflective or glass or something and we move our head around, we really expect light to behave in a certain way. We know what the reflection should look like. And these radiance fields are able to model them. So what you're able to do is you have like this individual radiance, the color of which you're looking at from a specific view angle. And it's just contained inside of a field, kind of like. Like a soccer field or football field, if you will. That's that the field is just what the method is being contained inside of.
- [03:17]
B
Okay, that makes sense. So really it's. So let me probably. I might have some questions that are way off because I'm obviously learning as well. I think that's the best part about this show is that when you say kind of the, the. The field or the. The field of vision. Right. So that makes total sense. I understand that. But the. So within creating, for example, a radiance field is the light and the angle and the refraction of that kind of the most important. What is like the most important part that makes it, quote unquote, a radiance field.
- [03:46]
A
So basically, yeah, it's just like trying to understand from a given viewing direction or viewing angle what should this actual resulting image look like? What would be the most reflective of the real world of that actual moment in time? Just like how we freeze a normal 2D image with our camera and it's able to, you know, capture everything in a 2D slice. It's kind of doing that same thing. But now we're removing composition as part of this, this thing. And now it's the entirety of the moment. It's a full 360. You can move around, you can explore everything. It's really just an entirety of a moment in time as opposed to just one specific viewing direction that's captured now.
- [04:21]
B
Makes total sense. So you mentioned two terms. You said Gaussian Splats as well as nerfs. High level. What are the difference between the both and what does it matter?
- [04:31]
A
Yeah, so basically nerfs are kind of the thing that kicked off these, this radiance field boom back in 2020. So it's a fairly new concept. And basically nerfs use something called ray casting, where essentially if you can imagine like a invisible ray extending from like a camera lens and basically what it will NERF does, it takes individual sample points along different points of that race. If you imagine like a fishing line or something being casted out and you're taking individual points along that fishing ray fishing line, which basically say like, what color is this given point from this viewing direction? And it does that for every single viewing direction from every single point in a 3D space. So when NERF first came out, it took obviously a very long time to complete this. This is, you know, operations on the order of like multiple billions. And so this is not something that was being done in real time by any means. But that really changed in 2022, where Nvidia put out a method called instant NGP, otherwise known as instant NERF. And that dropped the training time of NERFs down from, you know, over 24 hours to just a handful of seconds to a couple of minutes. And this is what really started to allow for the consumer and enterprise adoption of this technology. And that was pretty much that, you know, nerfs were like the state of the art for, for a While up until mid 2023, where we saw the publication of a paper called 3D Gaussian Splatting for Real Time Radiance Field rendering. And basically it changes out a few different primitives within the radiance field. But you're still necessary, you're still getting a radiance field at the end, still, still a very lifelike 3D structure. And basically instead of taking individual points samples along array, what Gaussian spotting does is essentially you can imagine a Gaussian as kind of like a 3D ellipsoid or like football shape that has varying like, like density and different, you know, properties to it. And if you could imagine like a Jackson Pollock painting, for instance, where it's got a lot of like paint splatters on a 2D canvas, right? But if you could imagine a computer instead of, you know, Jackson Pollock doing the painting, and instead of when, when they're kind of throwing the paint onto this 2D canvas, if the resulting image looked perfect from every single angle, now you're starting to get a sense of it literally is splatting these, these three dimensional Gaussians onto a 2D canvas which is essentially making up your viewing angle from say a 2D screen. So you will get potentially, you know, hundreds of thousands to millions of Gaussians that are representing a very lifelike 3D space in the end. But Gaussian spotting has a lot of really great properties that come with it compared to nerfs, at least for right now. And one of these things is very real time rendering rates whereby you are able to get rendering rates into the several hundreds of frames per second with Gaussian splatting. And it features something which, what's called an explicit representation and uses rasterization, which is something that's more in line with the traditional graphics pipeline. So we have a lot more things that are geared towards using Gaussian splatting right now compared to NERFs, which are such a, like fundamentally new and novel, you know, approximation. And so we, we can begin to see how Gaussian splatting is making an impact right now because we have a lot of things that essentially are well suited to the, for the time being for to, to use Gauss and spotting in the real world.
- [07:47]
B
So makes total sense. I mean, and I, that was all, a lot of that was new to me. So it's interesting that you, you know, I think you said it was ray something, but I think about graphics card and the traditional kind of visualization of Nvidia's graphic card, right where it's like ray tracing, right? It's like trying to track the light. So it makes total sense like how you said it's GPU dependent, but it's lowered the time from you know, 24 hours or days or longer down into, I mean obviously depends how many frames, but you know, how GPU dependent is the process of Gaussian spying or using nerfs. Like how, how dependent is it on the gpu?
- [08:23]
A
It's extremely dependent on the gpu. You know, there's really only a handful of ways to do it outside of like the, the Nvidia ecosystem. And it makes it a lot easier to, to work with these representations when you're able to really like fully leverage, you know, CUDA and the benefits of the Nvidia ecosystem. But like you were mentioning, you know, a lot of the, the work that's gone into this over since it's been, they've been released has really been able to make it scalable for larger scale use cases ranging from very large enterprise use cases to individual consumers as well, so it's really been able to service the whole gamut of use cases here.
- [09:06]
B
I love that. So, I mean, we're going to be showing some repetitive examples and Michael's going to be walking us through that. So hang on, we're going to get to that, but we're trying to kind of set the baseline because this is all about reshaping workflows. Right. So question for you, Michael, is, I mean, obviously there's probably more than we can talk about in the time allotted in terms of use cases that you've seen. But maybe highlight, let's say two industry use cases where radiance fields have picked up in popularity. Maybe describe, you know, kind of the, the workflow and kind of the, I won't say the unintended consequences. Not the right word, but like the, the benefit. Right. Is it a time savings? Is it a brand new thing that's never been able to been done before and now it's possible. Maybe pick maybe your top two or maybe the two most popular and we'll start with, we'll start with that.
- [09:56]
A
Sure, yeah. And I think it's, it's kind of an all of the above situation whereby with these, you know, radiance field representations, what's really incredible with them is that there is no barrier to entry in terms of the capture device. So anything that's able to take a still image is able to create the data set necessary for the reconstruction of a radiance field. And so to kind of illustrate two of the examples that are been pretty popular lately, the first one I'll take a look at is construction. And basically what, what's enabling people to do on site is they're able to really document accurately the specific state of the space. And so when people are doing site walks or site visits or being able to translate the state of their construction site to their stakeholders, essentially what they're able to do is with just, you know, a normal camera system, they can walk around, you know, take images of a given space, reconstruct that and then start to share it to their, their coworkers or be able to provide either documentation or annotations with this. And this is something that's not really been possible before because, you know, previously take a look at another technology like photogrammetry, for instance, which might really struggle with some of these thin structures and highly reflective objects. Radiance fields do not have this, this issue whereby they're able to reconstruct the entirety of the world without regards to if something is really reflective or has like kind of like a bit of a thin structure. So construction is one of the areas that you're seeing a lot of adoption of this technology. And another place that we're starting to see this technology become really powerful is in autonomous vehicle simulation and to, you know, essentially train self driving cars. Because you're able to get a very lifelike approximation of the physical space in the world. What these car companies such as, you know, Nvidia as well as Wave and Applied Intuition are doing is essentially they're taking a bunch of normal 2D images that they're capturing on say like some of their self driving cars that are out in the world. And they take these images and they reconstruct it into a very, very large radiance field. And basically once they have that, they begin to simulate it and they take the these representations and they, they create these situations which are known as like long tail scenarios. Things that don't necessarily happen often in the real world but are very important for us to be able to simulate and have an understanding of what the proper thing to do when we encounter that would be. And so they take these radiance fields and they drive an infinite amount of synthetic cars and run an infinite amount of synthetic scenarios. And they're using these to help simulate the actual real world. Because this is the closest approximation we have right now to being able to reconstruct the world. And so we're seeing a lot of interest now in the world of simulation for these, these different radiance field applications.
- [12:43]
B
Hey Michael, that's really interesting. I mean in, in terms of radiance fields I've, I mean I've heard about the use cases, you know, within construction. It makes perfect sense, right? Being able to go out, you know, use your phone, take a bunch of static images, stitch them together in a radiance field, or being able to go out and kind of take video, slicing those images up makes perfect sense to create kind of that 3D representation of the site without being on the site. But the self driving car, I mean kind of just blew my mind because one, I didn't know that. Two is, it's very interesting. I mean you mentioned the word long tail. I want to get into that in a second. But with explain to me how I get the idea of, you know, hey, we have a camera on top of the car, we're going to take it, we'll slice up the frames, we make a really long scene. But then how is AI then going in and creating a simulation, synthetic simulation inside of a radiance Field. That's really interesting.
- [13:38]
A
Yeah. So if you take a look at companies like Wave for instance, which just announced the release of their Gaia 2, which essentially there's like their Generative Engine. And another part that Wave has is something called like Prism, which is essentially a simulator that uses Gaussian splatting. And what basically they're able to do is that they have this, this general reconstruction of the world or of say a given space. And they're able to then place, you know, say like a synthetic car into this, this, this track. And they're able to, you know, say we know that there's a situation that we might not encounter that, that often they're able to introduce that and replicate that over and over and over again. And they're able to supplement it with some of these things like Gaia for instance, to introduce more of the generative side of things. But there are also companies like Nvidia who are releasing World foundation models like Cosmos, which are, you know, really tuned for physical AI, which is kind of like the perfect application for these radiance fields, whereby essentially if they have these say, you know, reconstructions, they have these like camera, these 2D images, they can start to enhance or enlarge and out paint some of the places that they are capturing to make larger and larger and more complete environments, things that are more closely resembling the real world. And this is why some of these World foundation models are so valuable is they're able to, you know, essentially create this temporally consistent and plausible generative video that can then be turned into a 3D world essentially and be used for, for simulation based purposes. And so this is where we see a lot of the work of the generative 2D starting to spill over and actually create value in the world of simulated 3D.
- [15:19]
B
That's amazing. I mean, and it makes sense. And I played with the, the Nvidia's kind of, I mean it was the image to, or text to image and then the image to video models. And you're right, it is creating basically a synthetic kind of 2D experience. I tried to create like scenes with people. Not real good, real good at creating like a scene with a robot. Like, you know what I mean? Very good at that. But you kind of say long tail. Like it kind of leads me to the question is that, you know, where have you seen, for example, where do you would you see kind of a smaller short tail radiance field being used versus like a long tail. Like the long tail being obviously driving many hours, et cetera, but like, where is. Like, where's the practical application of short versus long?
- [16:03]
A
Yeah, well, so say, take an example where, say, you know, you are driving and you turn down like an alleyway or something, and you, you start going like 5 or so feet and all of a sudden like a car that's in front of you starts like going in reverse back towards you. It's like this, this is not a scenario that you would encounter very often at all in the physical world, but it's a very important one to say, like, okay, if another car is going the wrong direction at me in this, you know, very niche environment and is, you know, all of a sudden happening like, like, what's the best way to respond to this? Is this something that I should also, you know, go in reverse and try to back out of the alleyway into perhaps oncoming traffic where I can't necessarily see, like, what's the best way to actually, you know, handle these scenarios. And it's much easier to create these environments in a synthetic environment as opposed to essentially going out in the real world with a real car and starting to just, you know, this is something that's a lot harder to scale because, you know, and anyone who's like driven, you know, knows that at any given moment some random thing can happen that, you know, you might be getting caught off guard by. And so it's much easier to, to feed these data points to a computer upfront as opposed to letting them encounter them for the first time in the real world.
- [17:11]
B
Makes total sense. I mean, I mean, it's so interesting. I mean, I, I love, I just, I don't know, it's just, it's. Honestly, I hate to say this and I'm, I'm. But I probably upset other guests. By far the most fascinating topic that I think we've had, right? Because I mean, the application really hasn't been that long. You said 2020 ish, but really accelerated 2022. So like, I mean, let's do the math. I mean, it's not. We're barely in a 2025, so we're like a two year time horizon. It sounds like a lot of things. And it's so, and I think it's so accessible in the sense that, you know, it's literally your phone, right? You're not taking huge like LiDAR systems or, you know, all of this type of stuff to go around and try to capture, you know, stuff. It's literally your phone. But what I like to do is I know that you've kind of queued Up a couple of examples and I'd love for you to kind of share your screen and kind of walk through the two examples. The first one, you know, being kind of the, the coat of arms, kind of the alcohol, the Renaissance room. Right, starting with that because want to kind of talk through the workflow, right? Like what, what camera did you use? Like what tools are you using? Like you know, what is the hardware rig you're using? What are some of the best practices? Because the whole idea of this show is to give people the ability, you know, obviously equipped with the proper Dell Pro Max products as well as Nvidia RTX GPUs, but to be able to go out and say hey, I'm going to go give this a shot. So you're the expert, school us.
- [18:38]
A
Awesome. So I have right here a currently free to use platform called Post shot whereby you can just, you know, start to train your images. And we can see here, this looks like a normal 2D image, right? This looks like a normal room that has some nice light coming into it, got a lot of detail. But it's actually, this is actually not a 2D image and it's not a video either. It's actually a fully interactive and you know, very lifelike 3D model. And we get a sense as to what we can actually reconstruct and the level of detail that we can start to create from, from these different radiance field technologies. And so this capture actually was taken by one of my friends whose name is Christoph Schindler, who is based out of Germany I believe. But it's pretty stunning the amount of fidelity that you can get inside of one of these captures.
- [19:25]
B
I mean that's incredible. So obviously a bunch of questions. Let's start with the first question. Obviously a real place, how many images were taken to create this radiance field?
- [19:37]
A
So this I believe was about 600 images that he took of the room.
- [19:41]
B
Okay. So best practices for people who aren't super like great with the camera, which is me and probably my 10 year old daughter, is like when you're trying to construct a radiance field, right. Maybe if you zoom in, if you don't mind, maybe zoom in on one of the like coat of arms on the side might be, might be a good example. Okay, right, yeah, right there's perfect. Yeah, that's perfect. So when you're looking at that is the idea to create a bit of an overlap between each shot where if you go kind of like take above the door, right? Do you want to get just Kind of the edge and then the next photo over is like. So basically what are the best photo practices to make sure you're putting together? You're, you're. Even though you're taking 600 pictures or how many ever, you're grabbing the right 600 pictures.
- [20:26]
A
Yeah. So Radens Fields love more than anything else two, two things. They really love having very sharp frames, like where everything is in focus, so that, you know, essentially they model whatever you capture. So the more things that are sharp, that are in focus, the higher quality typically the results will be. And the second thing that Radiance Fields also really love is something called parallax, whereby you can start to introduce this, this movement where essentially, you know, you, you can create parallax from say like creating like, like an orbit around a scene or just like you were mentioning as well, having some level of overlap from one image to the next so that, that the system essentially can start to understand where in a given 3D space that these cameras are. And if I actually toggle on here, we can actually start to see that the actual cameras that were used in this reconstruction and where the actual images were coming from. So we get a sense as to, you know, you could see that a lot of them are occurring on, alongside the edges and are looking into the scene itself as well. As, you know, you can see there are a couple as well that are actually in the, the, the middle of the room too. But getting it. Being able to get this like, larger amount of coverage of the space is very important because essentially these Radiance Field pipelines will model whatever they see. And you know, the better. It's kind of like, you know, when you go to edit a 2D video and if you introduce just a bunch of like video footage where it's really shaky, it's really blurry, the audio is bad. You know, you're gonna have a very tough time getting something high quality out of it. You, you wouldn't see this in like, say like a professional film, but it's the same exact thing here where you, you know, the higher quality you're able to introduce upfront, the, the easier time you'll have at the end of being able to reconstruct this. And here we get a sense as to like, what fidelity level is possible, you know, using this, this technology.
- [22:14]
B
It's, I mean, it looks honestly like I could like walk inside the room. I mean, it's fantastic. Like how, in terms of how big this room is, is there kind of a. So you said more images, the better and you know, my, my background a little bit with AI, right. And I, when you're training, for example, you know, I, I've done some animation loras and stuff where I'm taking, you know, prior, you know, prior animations, trying for character consistency, et cetera. And there's actually more of a sweet spot. Like it's not just necessarily more, it's kind of the differing of the images. Right. And if you add too much, it kind of muddies it in. Is there a rule of thumb just hey, take as many images as you possibly can from as many rooms. Like is there, you know, a long room like this? I have no idea how big it is, but let's say it's 50 foot wide or you know, 50 foot long and then say 20ft wide approximately. Is there kind of a formula that you should use or is it really just make sure you capture everything, different viewpoints, doesn't matter how many there are.
- [23:14]
A
There'S not really a best practice that's known yet, you know, but I would say that like normally around 600 images is probably like the most that you would really require for, for like most spaces, you know, definitely when you start to really create very expansive captures or if you're going to be doing say multiple rooms, then you might be segmenting out each dataset into its own reconstruction. But typically I would say I personally try to shoot, you know, say if I'm trying to capture a person, for instance, I might only use like 120 frames and that could be sliced from a video as well too. But typically I'd say that the sweet spot probably is between 300 to 600, depending on the size of the area in which you're trying to capture. But more than anything, introducing sharp frames is the most important thing. And you know, being able to shoot with a very high shutter speed, you know, more than a 500th of a second is greatly valuable. And another thing too is that can be very helpful in this way when we're talking about like how many images to use is like what's the lens that you're using. And so if you're using a more wide angle lens, like say like a 14 millimeter lens, you're able to bring in a lot more data per image than you would if you were using a 50 millimeter, for instance. And there's a lot of other really nice things that you get when you're using some of these, like wider angle lens, such as like this thing that's called like your hypo hyper focal distance where you can start to set your, your focus to infinity. This way you don't need to necessarily worry about like, oh, is, is the, the foreground going to be really sharp and the background's going to be totally blurry because you don't want that. You want to have something that really the entirety of the image is super, super sharp all the way from the foreground to the background. So I would say that the lens choice also makes a big difference, just like it does in normal 2D photography. But here you would want to prioritize more wide angle for your captures.
- [25:04]
B
So more as wide angle as you want. Very sharp and crisp photos. Be methodical in what you're taking. Make sure you capture everything. All right, let's say now we have our 600 picture data set. When you load them into, you know, into post shot, is it put them in in any order? Is it the order in which you took them? Do you do walls first? Like, how does that process work?
- [25:29]
A
So basically, once you've taken your complete data set of all the images, you can actually just drag and drop them into these, this post shot window here. There's actually a couple ways that you can go about this, but for. With different like, reconstruction softwares. But for post shot specifically, you can just take that image dataset. It does not need to be ordered in a specific, you know, way as long as it's just the specific images that you want to use and you just drag and drop it. And basically it will restart the actual camera tracking process, which I can try and see if I can restart it right now. We'll take a look at it. But see, so I'll start it here again. I just dropped in the folder and you can start to see that it's now starting to retrain everything and. Oh, looks like I'll just. So you can see now it's starting to reconstruct everything and it actually looks like it's dropped it on top of the original one. You can see here. Yeah, yeah. So that's why it looks the way that it does currently. But you can, you can get the sense as to like, you know, things still would be a bit more translucent until it begins to be, you know, completely trained at that point.
- [26:36]
B
Okay, is there any other, like, is there any other kind of setup or best practices if somebody wants to go out and capture, let's say inside their home as a radiance field to get started? Is there any other best practices?
- [26:48]
A
Definitely just being able to, you know, it really comes down to having sharp frames and and having parallax that. Those are definitely like the two most powerful things that you can do. It's better to just, you know, remove blurry images. Just don't, don't use them to delete them before you start to do the training. It's. It would not help. So it. I'd much rather have a smaller collection of very high quality images than more overall images for the reconstruction. And it's very cool. But you can also use, you can also reconstruct mirrors inside of radiance fields as well too, which is like a, kind of like a weird, you know, thing that, that some people like, might not consider for like interiors. But you can essentially create that as, as well. I'll see if I can pull up an example too.
- [27:36]
B
Hold on. I got it. So I gotta ask about this. So you're, let's say a bedroom. You're creating a mirror, but a mirror is reflective. So what would reflect in the mirror of a radiance field that. Would it be the person who captured it? It. Would it be like, what would it be? I'm curious.
- [27:52]
A
Yeah. So technically, if you are not careful with the way that you capture it, then you will appear inside of the reconstruction. But there are ways where essentially, here's an example where you can see, here's the mirror and I am not contained inside of it.
- [28:07]
B
Because you clearly ducked out, I think. Did you duck down? Because the way the camera angle went, I figured you ducked out.
- [28:14]
A
Yeah. I have some methods to avoid appearing in mirrors. Okay. In my captures. But you can see, you know, like a lot of these views here, you know, I am just nowhere to be found. And if you go into the other, the other side of this too, you know, we can start to see. Let's see. Yeah, so now we can see. Now we're in the mirror world of, of this capture.
- [28:34]
B
We're in the mirror. I love it. That's awesome.
- [28:37]
A
Yeah. But now you can see, you can fly back. And now we're back to the normal.
- [28:41]
B
Part, you know, So I mean, it's really even beyond. So Rayfield's even really beyond like, you know, the traditional manufacturing aco. Right. I mean, interior design, another one, like obviously very, you know, very relevant if you wanted to design, mock up. I mean, my wife is pushing me very hard to redo our house, you know, to show a client a potential design and be able to maybe move that into omniverse or other things. Right. Like maybe give me. Well, let's ask kind of two questions because I know that we're Kind of getting kind of towards the end, I mean, you've been so gracious with your time is what are, let's start with this, is that within radiance fields we kind of talked about 2020, 2022, the birth accelerating since 2022 till now. What are some of the big things within radiance fields that are either on the horizon or that you see on the horizon for kind of the space or say over the next year?
- [29:39]
A
Yeah, so now I would say that, you know, a lot of very diverse industries have begun to understand that we can reconstruct anything contained in the physical world in a very lifelike capacity. And so we're starting to see, you know, either both behind the scenes as well as starting to now trickle out into the public world some very, very large industries such as telecom as well as like, as you're mentioning, like real estate and interior design and AC geospatial, you know, disaster relief, where essentially we're able to now for the first time in like human history, be able to very quickly reconstruct the entire physical world, whatever that that thing or use case might be. So essentially over the next year I really do expect to continue to see more industry adoption across anyone who currently uses some form of 2D in order to actually visualize or pull out some level of information about the three dimensional space or could be able to convey information through this. So another industry that's really starting to embrace this technology is E Commerce, whereby, you know, if you were going to go to Amazon and you're going to buy something but you're not necessarily sure like what exactly am I getting here? And sometimes you don't know until it actually shows up at your door. But now with, with These different represent 3D scale representations such as Gaussian spotting, we're able to get a very lifelike approximation and be able to explore and actually see the, you know, fine details on, everywhere on, on this object so that we're able to make more informed buying and purchasing decisions. And so we're seeing a lot of these E commerce players either experiment with adopt or adopt this technology. So I think in the next year we're going to see a lot of these industries begin to use it. And then probably the next like three to five years we're going to start to see, at this point I am starting to believe that we're going to start to see a shift of imaging to evolve out of 2D and into a very lifelike 3D world because we now have the capability of reconstructing Everything in the world in 3D, including moments in time for people, you know, it's able to create.
- [31:44]
B
Yeah, why would you want, why wouldn't you want that?
- [31:46]
A
Yes, exactly. Yes. And I think for, for me, like one of the core purposes of imaging is to be able to translate like human experiences to, to one another. And now we really have a very straightforward pipeline to creating this where we can really step back into a moment in time for any use case or in any application. And so I think that that's where, you know, in the somewhat medium term we're going to start to see this technology really start to make a dramatic impact on our everyday lives outside of just the industrial applications.
- [32:14]
B
Yeah, I mean, and you're right, the application, I mean if you have the ability to make it 3D, the application is huge. Right. So it makes total sense. I didn't know about Amazon and others, but I mean, it makes sense. I mean there's been a lot of times where I have, well, not me necessarily, but my wife has bought something for my daughter or whatever and then it looked one way and then it came. But it might have been way small, like, you know what I mean? Like way smaller than we thought. And I'll, I'll, I'll tell you, it was really funny. Is she wanted like this like stuffed toy or whatever, right? And, but the images made it look like it was, you know, fairly sizable. But then when it came, it was like kind of like the cap to my, to my, my webcam, right. And Radiance Field, you would have known because it could have been in a physical space inside the box, all these things. So you would have known. I've been like, oh, that looks little, little sketch. It looks not quite right. So, you know, this has been fantastic. I, I guess the one last thing that I want to hit upon. We kind of talked about the future, We've kind of talked about the past. You know what I think the, the where because a lot of, a lot of folks within this show, right? Like are using traditional, what I would call ISV workflows, right? Whether that's media, entertainment, engineering, you know, oil and gas with ISVs like think Siemens, Ansys, Autodesk, all these things, right? Where do you see those ISVs picking up kind of Radiance Field injections into their ISV workflows. Have you seen that yet? Is that coming, Is that like the next horizon or is it kind of going to be a standalone thing that will ultimately get ported into one of those applications?
- [33:57]
A
No, I think that it's it's really interesting some of the more industrial applications of this technology because, you know, for me, while we started to take a look at a lot of the really lifelike visuals that you can get from a radiance field, I actually look at the visuals as the most foundational layer about what you can do with this technology. And once you have this lifelike detail, you can start to embed information in a very spatial way where you can, you know, depending on like what your use case is like, if you are doing say maintenance on like an oil rig in like a very remote destination, you know, you can create essentially what, what's known as like a four dimensional time lapse of this, this oil rig and you're able to actually create annotations or notes or you know, be able to flag things to people to actually pull out some level of like business intelligence for the, these use cases. So I see a lot of more heavy use of AI in the future with this technology because one of the examples that I'll give with this is that you can actually pair LLMs with a radiance field. So something like ChatGPT, right? Where you can start to say, you can start to have an LLM like ChatGPT, understand what something is contained inside of it. So you could say like, okay, you know, if I'm on a construction site, flag me areas that might be safety hazards right now and it can go, and it can automate the process, like for people to do this to actually make sure that things are moving to schedule or make sure that anything that could be taken advantage of in the future will be able to do so. So I see a lot of like very heavy crossover into the world of, I guess you could call like the industrial applications of this technology.
- [35:36]
B
That's great. So like I said, I hate to say it, probably one of my favorite episodes. I love it. This has been so fascinating because it's so different than what we normally do. So, you know, Michael, you've been very gracious with your time. Take a second, tell, you know, everyone who's listening and watching, you know where they can find you, your website, you know, any other LinkedIn, any other details. So if they want to connect with you, they can.
- [36:02]
A
Yeah, so you can Find me on LinkedIn. My name is just Michael Rublef there, but I primarily post on my website, which is just radiancefields. Com. I try and post daily now given the pace of this advancements. And yeah, I continue to provide consulting services for businesses that want to either experiment with or begin to adopt or learn more about this emerging set of technology. So if you have any questions at all, you know, feel free to reach out to me either through LinkedIn or my website. Or if you just want to say hi, you know, definitely just reach out and we can always have a conversation. So, yeah, I appreciate it.
- [36:35]
B
Yeah, I would highly recommend it. Michael's very educated guy. I got to sit down with him even more at gtc and it was, it was a fantastic time. So, you know, with that being said, you know, kind of wrap up here, is that, you know, within the show, I mean, within the whole scope of the show, right, Is that we're trying to show you interesting new workflows that Dell Pro Max and Nvidia are enabling that can really empower, reshape and accelerate the way that you do day to day, or the way that you look at the world or the way that you look at your business.
- [37:04]
A
Right?
- [37:05]
B
And Radiant fields being one of them. Think about it like five years ago, would you have ever thought you could take your camera and recreate without very expensive equipment? You know, hours and hours and hours and hours and hours and years of rendering time, being able to do it almost in the fly with 600 photos, right? So now that you know about Rainsfields, think about your business. Think about how could you benefit from this, you know, what could you do with this technology? And I would encourage you to go check out, post shot, take some photos and play and learn, experiment, ideate, because that's the only way that you're going to be able to stay up with this technology is to be able to go do and try. And that's, at the end of the day, what this show is really about. So with that being said, it's Logan once again. Hopefully you're not sick of me. And this is reshaping workflow. Signing out. Keep your workflows running on Dell, Promax and Nvidia and I'll see you on the next one. Do what you want.
- [38:18]
A
This podcast was produced in partnership with Amaze Media Labs.